Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: Tim Gray on October 13, 2007, 09:06:38 am
-
http://www.maxmax.com/nikon_d200hr.htm (http://www.maxmax.com/nikon_d200hr.htm)
Link from the Online Photographer
http://theonlinephotographer.com/the_onlin...blog_index.html (http://theonlinephotographer.com/the_online_photographer/blog_index.html) of Oct 12.
As Mike Johnston says "live dangerously".
I've seen it mentioned before that there are effective software options to remove/reduce the resulting increase in moire - what works best?
-
well I would do it.
the results I've seen from the samples are similar to what I get from a MFDB. It does make a difference in a print. Whether it's worth that amount of money to you is a different matter, that's all.
Treat it as a specialist mod like the IR or UV mods
personally if I had a 5D I would do it, but they don't seem to offer the mod for 1DsMkII
-
The results from the D200 mod do look worthwhile from what they have on their website. The results for the 5D less than impressive, when looking at the actual raw files in different raw apps that I use regularly. The cost is not a big deal, but if I see a nearly miniscule to no difference at 100%, then what in the world does a print comparison add. So maybe the price of the mod for the 5D isn't worth it. If I don't see much of a difference then comparing to a digital back is of little value.
The jpeg samples they show do look very different. But that is not the case looking at the raw files. And in the end they are trying to sell a service here. If someone is trying to squeeze one more percent of detail out of an image through some hardware mod, then maybe they are into the gear more than the art of photography.
And one bit of truth, color moire can be reduced through software, luminance moire patterns are hard as hell to get rid of. I would like to spend less time in front of the computer not more.
[a href=\"index.php?act=findpost&pid=145750\"][{POST_SNAPBACK}][/a]
actually you'll be surprised at the difference a print makes vs computer screen, even at 100%.
As we speak now I'm slowly uprezzing a file and sharpening it for 4ft x6ft to 240 dpi.
it took alot of trial and error of printing on the epson and comparing to the print to realise that what's on a screen can translate differently in print, in terms of what the human eye perceives as photographic or not (when is sharpening too much sharpening to the point of looking flatly digital?)
Talking about sharpening an AA-ed image, when I uprez my 1DsMkII file and I sharpen it I find that it's impossible to bring back minute detail that's soft through the obvious AA filtering in the first place.
You can sharpen an image to bring back local contrast but then when details are mushed in the first place,sharpening it later leaves pretty half-assed details. And using local contrast sharpening before uprezzing the image just leaves a degraded image.
When you look at images 100% on a computer screen you're seeing a rendering of square pixels forming an image, and when you print it out you're looking at dots. Too many people make comments based on looking at computer screen images, would you consider printing the pictures out to see if you can tell a difference?
I think at small print sizes you can't tell much difference but there was this image I shot on a H39 and I always thought it was unspectacular, until one day I had to blow it up to 5 ft and suddenly the picture seemed to pop out alot more compared to my 1DsMkII file, details rendered differently although on the computer screen you would be drawn to the 1DsMkII file instead
I'm weighing in on this topic because I personally feel that images do benefit from a lack of AA filter.
I shot with a Kodak Pro Back quite awhile back and even then I already felt that the images rendered at the optimal ISO was providing me images that had finer details and a more photographic, dimensional look than my MkII which, as much as I love it, does seem like its images have been shot through some soft cloth sometimes in terms of rendering details.
People are willing to pay alot of money to buy a zeiss lens simply for subtle gains in terms of rendering of an image, so I don't see why there's so much negativity regarding the service being offered.
once again it's abit of a specialist service, and not suited for everyone I'm sure.
Personally I would do it in a flash if only they offered it for my camera.
I'm sure that the company is willing to provide more samples should you email them
Plus there's a forum member who's had his 5D modded and he seems happy with it. Perhaps you can ask him for his opinion on it.
-
also I wonder if removing the AA filter will help with the problem that occurs due to diffraction
Sometimes I have to shoot F16 and trust me your images ain't gonna look pretty at F16 on a full frame sensor...details are absolute mush.
but ahhh I'll just think hypothetically for now.
-
http://www.maxmax.com/nikon_d200hr.htm (http://www.maxmax.com/nikon_d200hr.htm)
Link from the Online Photographer
http://theonlinephotographer.com/the_onlin...blog_index.html (http://theonlinephotographer.com/the_online_photographer/blog_index.html) of Oct 12.
As Mike Johnston says "live dangerously".
I've seen it mentioned before that there are effective software options to remove/reduce the resulting increase in moire - what works best?
[a href=\"index.php?act=findpost&pid=145701\"][{POST_SNAPBACK}][/a]
The democarcy of these forums is that everyone's opinion is equal. But not all of them are correct. With digital imaging, we are not entiltled to infinite resolution; the attainable resolution is limited by the sampling frequency of the sensor, not the lens. In fact, if you ignore this inconvenient truth, you get aliasing. Aliasing of periodic elements,such as fabrics, are called Moire patterns. The image is distorted, with high frequency details in the wrong location. There is no such thing as color Moire patterns. The sampling problem for Bayer sensors is complicated by the different sampling frequencies of the red, blue and green detector elements. Once this distortion is observed in an image, there is no sure way to remove it unless you know what the image is supposed to look like. In other words there can sometimes be mitigation of the effect but there is no general method to remove aliasing.
The AA fiter is there to attenuate high frequency information beyond that which can be faithfully recorded by the sensor. It is digital imaging 101 to roll off the high frequency information above the Nyquist frequency of the sensor. This is THE standard approach in digital signal processing. No amount of voting or opinion is going to change this. One goes naked at their own peril. In digital photography, the roll off of the high frequency information is done by blurring the image before it reaches the detector with an AA filter. In paractice, AA filters may not be perfect. The goldilocks effect would be not too weak to allow aliasing and not too strong to remove any detail. With the AA filter it is generally helpful to apply a small amount of sharpening during raw conversion.
Arguing, for example, that the Leica M8 is better because it lacks an AA filter, is a hypothesis. There is no proof. It might be some other element in their proprietary development. To prove it is the lack of an AA filter one needs to make measurements with an M8, with and without an AA filter. Measurements, not pictures shot out your porch window that are later displayed as jpegs on a computer monitor.
Sorry fot the rant, but this same subject has just been explored in Cameras, Lenses and Shooting gear.
-
The democarcy of these forums is that everyone's opinion is equal. But not all of them are correct. With digital imaging, we are not entiltled to infinite resolution; the attainable resolution is limited by the sampling frequency of the sensor, not the lens. In fact, if you ignore this inconvenient truth, you get aliasing. Aliasing of periodic elements,such as fabrics, are called Moire patterns. The image is distorted, with high frequency details in the wrong location. There is no such thing as color Moire patterns. The sampling problem for Bayer sensors is complicated by the different sampling frequencies of the red, blue and green detector elements. Once this distortion is observed in an image, there is no sure way to remove it unless you know what the image is supposed to look like. In other words there can sometimes be mitigation of the effect but there is no general method to remove aliasing.
The AA fiter is there to attenuate high frequency information beyond that which can be faithfully recorded by the sensor. It is digital imaging 101 to roll off the high frequency information above the Nyquist frequency of the sensor. This is THE standard approach in digital signal processing. No amount of voting or opinion is going to change this. One goes naked at their own peril. In digital photography, the roll off of the high frequency information is done by blurring the image before it reaches the detector with an AA filter. In paractice, AA filters may not be perfect. The goldilocks effect would be not too weak to allow aliasing and not too strong to remove any detail. With the AA filter it is generally helpful to apply a small amount of sharpening during raw conversion.
Arguing, for example, that the Leica M8 is better because it lacks an AA filter, is a hypothesis. There is no proof. It might be some other element in their proprietary development. To prove it is the lack of an AA filter one needs to make measurements with an M8, with and without an AA filter. Measurements, not pictures shot out your porch window that are later displayed as jpegs on a computer monitor.
Sorry fot the rant, but this same subject has just been explored in Cameras, Lenses and Shooting gear.
[a href=\"index.php?act=findpost&pid=145758\"][{POST_SNAPBACK}][/a]
the problem with applying sharpening during RAW conversion is that it leads to a degrading of the edges of objects in the image when the image is uprezzed.
I tried uprezzing from an image that was lightly sharpened during RAW conversion, and I tried uprezzing from an image that was unsharpened during RAW conversion,
and then did 10 percent bicubic increments (which btw does make a difference as opposed to uprezzing to full resolution directly, even with CS3), and then sharpening the final resolution images after that and the initially unsharpened images hold up to sharpening better in terms of smoothness and photographic quality, whereas the initially sharpened image tends to become chunky at the edges (pixellation and heavy contrast).
The difference to me is pretty big but I think to most people it won't make much of a difference and most people don't print too big anyway, but for all the talk about Nyquist and whatever technical issues...(sorry my mind blanks out at the mention of such words), at the end of the day the final product in front of me speaks for itself.
-
I've had my modified 5D for 9 months now without a AA filter. I find very little moiré and the Image Quality after sharpening (prints) is improved to my eye. I have no regrets and the $450 was worth the improvement (for me). This is a subjective/empirical evaluation of course. Those of you with a AA filter can theorize all you want but there is a real world improvement and my conclusion is; in the current high megapixel sensors for landscapes and nature , the AA filter is not necessary it is a hindrance. About 7500 frames to date w/o the AA filter.
Marc
-
You may want to look at this thread for more discussion on this topic
http://luminous-landscape.com/forum/index....opic=20081&st=0 (http://luminous-landscape.com/forum/index.php?showtopic=20081&st=0)
Would someone like to perhaps enlighten me on why most medium format backs seem to forgo the AA filter?
and medium format backs are held up as the holy grail of image quality
-
You may want to look at this thread for more discussion on this topic
http://luminous-landscape.com/forum/index....opic=20081&st=0 (http://luminous-landscape.com/forum/index.php?showtopic=20081&st=0)
Would someone like to perhaps enlighten me on why most medium format backs seem to forgo the AA filter?
and medium format backs are held up as the holy grail of image quality
[a href=\"index.php?act=findpost&pid=145765\"][{POST_SNAPBACK}][/a]
Only my opinion, MFDB have high megapixel sensors and IQ is paramount to their end users. MFDB are not general purpose cameras but specialized tools.
Marc
-
You may want to look at this thread for more discussion on this topic
http://luminous-landscape.com/forum/index....opic=20081&st=0 (http://luminous-landscape.com/forum/index.php?showtopic=20081&st=0)
Would someone like to perhaps enlighten me on why most medium format backs seem to forgo the AA filter?
and medium format backs are held up as the holy grail of image quality
[a href=\"index.php?act=findpost&pid=145765\"][{POST_SNAPBACK}][/a]
Are you suggesting that the lack of an AA filter on medium format backs proves something? We don't know what a MF camera could do with a properly designed AA filter. Maybe it is left off to maximize profit . It's all speculation. Is the emperor dressed or naked? We do know from theory and practice, that distortions can be observed in cameras without an AA filte. We are just arguing about exactly where the line is to be drawn. Maybe if you get a sharper lens you will be less thrilled with your AA-less camera.
-
Are you suggesting that the lack of an AA filter on medium format backs proves something? We don't know what a MF camera could do with a properly designed AA filter. Maybe it is left off to maximize profit . It's all speculation. Is the emperor dressed or naked? We do know from theory and practice, that distortions can be observed in cameras without an AA filte. We are just arguing about exactly where the line is to be drawn. Maybe if you get a sharper lens you will be less thrilled with your AA-less camera.
[a href=\"index.php?act=findpost&pid=145804\"][{POST_SNAPBACK}][/a]
yes I am suggesting it proves something.
You can't get sharper than a Rodenstock HR lens that people use for architectural purposes with their medium format digital backs.
I'm sure they come across lots of air conditioning vents too.
regarding theory and practise I don't know why no one is talking about experience with actual prints rather than 100% on a computer screen
looks like no one believes me when I say that it's different...
-
I just got the November issue of Photo Techniques, in which Mark Dubovoy compares several up-rezzing software options. It's a well-written article with many comparisons of final results.
He ends his article with a discussion of capture devices (cameras) and notes that up-rezzing photos taken with the Leica M8 are better than those taken with the 1Ds MkII, in part because of the lack of an AA filter in the M8.
-
I just got the November issue of Photo Techniques, in which Mark Dubovoy compares several up-rezzing software options. It's a well-written article with many comparisons of final results.
[a href=\"index.php?act=findpost&pid=145825\"][{POST_SNAPBACK}][/a]
Off-topic, but could you provide a Cliff's Notes about the article? It's been a while since I've tested the different uprezzers.
-
Marc,
I appreciate the first hand account. Any chance that you could provide me a raw that shows the uber sharpness and an appealing image to boot. I am genuinely interested. The samples on maxmax just uninspire. I'd be happy to give you my FTP info for it.
[a href=\"index.php?act=findpost&pid=145815\"][{POST_SNAPBACK}][/a]
Yes of course whats your email address?
Marc
-
Well, certainly in Michael's first comments on the 1ds3 the existence of the AA filter features prominently.
I'm more curious than ever if it's even possible for an aftermarket removal of the filter (particularly given the anti dust functionality).
-
I'd not recommend it.... It's there for a reason - do you think Canon would like to get sharper images and reduce their camera's cost by removing it if they thought they could get away with it. You can criticise Canon all you like for not having a mirror lock up button or whatever, but when it comes to engineering digital cameras, in terms of performance and image quality, they certainly know what they're doing.
As for the medium format companies, I'd reckon that a defect free AA filter of the size they'd need could be rather expensive :-)
Graeme
Well, certainly in Michael's first comments on the 1ds3 the existence of the AA filter features prominently.
I'm more curious than ever if it's even possible for an aftermarket removal of the filter (particularly given the anti dust functionality).
[a href=\"index.php?act=findpost&pid=147035\"][{POST_SNAPBACK}][/a]
-
For years unbiased people know Nikon models, especially the D70/D70s, use very weak AA filters, thus generally the per pixel sharpness is higher out of Nikon bodies. My D2x provides around 30% more true resolution than my 5D, both with AA filter, and with over 90% of my work still with 4x5 and 6x7 film I am sure I know what I am talking about.
Like every manufacture, Canon, or anybody else, does not know everything they are doing. The 1DIII AF problem is an example. I shoot and love my 5D more than the Nikon, mostly because I spent more on higher grade Canon lenses, but I am amazed ingorant people talking about the "surperior Canon technology" with their heads burried in the sand. For example I see repeatedly people are saying Nikon D3 steal Canon's 1DIII/1DsIII battery, while in fact Nikon started using the same battery for 3 years by now and the fact is just other way around.
In my native country China they have created special words to describe Canon images for landscape works, "ROU", which means muddy looking which is mostly from the AA filter. My 5D does not have such problem since the filter has been removed and I had never problems with it. I am speaking to people with large format film experiences and understand what a 30x40 fine print would look like, not the MP3 generation.
You can criticise Canon all you like for not having a mirror lock up button or whatever, but when it comes to engineering digital cameras, in terms of performance and image quality, they certainly know what they're doing.
As for the medium format companies, I'd reckon that a defect free AA filter of the size they'd need could be rather expensive :-)
Graeme
[a href=\"index.php?act=findpost&pid=147050\"][{POST_SNAPBACK}][/a]
-
Sure Canon don't know everything, but both Canon and Nikon know that an optical low pass filter is a required element. You can debate back and forth on who's implementation is better, but as other factors effect the appearance of aliasing, like photosite fill-factor and microlenses (as does the MTF of the actual lens used), you can't easily do a "like for like" comparison. What I'd do (and indeed, this is what I do do) is get a large, high resolution zone plate test chart and shoot that. It shows you pretty much exactly what you need to know in terms of resolution and aliasing.
Graeme
-
I've seen it mentioned before that there are effective software options to remove/reduce the resulting increase in moire - what works best?[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=145701\")
No software can do a good job of removing this moire ([a href=\"http://en.wikipedia.org/wiki/Image:Moire_pattern_of_bricks_small.jpg]GNU Licensed Image, usage allowed[/url]):
(http://upload.wikimedia.org/wikipedia/commons/f/fb/Moire_pattern_of_bricks_small.jpg)
-
My D2x provides around 30% more true resolution than my 5D, both with AA filter, and with over 90% of my work still with 4x5 and 6x7 film I am sure I know what I am talking about.
[a href=\"index.php?act=findpost&pid=147054\"][{POST_SNAPBACK}][/a]
What do you mean by 30% more true resolution? 30% more lines per picture height or picture width?
According to dpreview tests of line charts, the D2X delivers around 20% more lines per picture height than the 8mp 20D.
Perhaps you are referring to typical resolution in the corners of the frame where cropped 35mm formats usually have an advantage.
-
Resolution is meaningless if it's corrupted by aliases.
Graeme
-
LEPING,
Where did you go to have your AA filter removed?
-
Resolution is meaningless if it's corrupted by aliases.
Graeme
[a href=\"index.php?act=findpost&pid=147072\"][{POST_SNAPBACK}][/a]
Note that LEPING wrote that the D2X provides about 30% more resolution than his 5D both with AA filter.
-
Resolution is meaningless if it's corrupted by aliases.
[a href=\"index.php?act=findpost&pid=147072\"][{POST_SNAPBACK}][/a]
The standard for measuring resolution, like lines per mm, comes from film which does not alias in a regular pattern. Therefore, no one ever safeguarded the term "resolution" against aliasing in grid-like captures. I think the definition of photographic resolution needs to be upgraded to "that which can be reproduced consistently with any random chance alignment of subject and photosites".
An aliased capture can capture alternating black and white lines on every line, but shift the alignment 1/2 the pixel pitch, and nothing is captured at all, except solid grey. This former is not "resolution".
-
That's why I use a nice linear sinusoidal zone plate to measure resolution and test demosaic algorithms.... It makes all this clear and plainly visible.
Graeme
-
That's a perfect image to show how the aliases fold back to a lower spatial frequency, and hence cannot now be removed by a blur, other than a blur so large that ruins the image and removes all the detail in it. There's no way known to repair an image like that and remove it's aliases.
Graeme
No software can do a good job of removing this moire (GNU Licensed Image, usage allowed (http://en.wikipedia.org/wiki/Image:Moire_pattern_of_bricks_small.jpg)):
(http://upload.wikimedia.org/wikipedia/commons/f/fb/Moire_pattern_of_bricks_small.jpg)
[a href=\"index.php?act=findpost&pid=147070\"][{POST_SNAPBACK}][/a]
-
Resolution is meaningless if it's corrupted by aliases.
Graeme
[a href=\"index.php?act=findpost&pid=147072\"][{POST_SNAPBACK}][/a]
blahblahblahblah
Let's wait to see more people upgrade their cameras and hopefully post
I've been in touch with Dan Llewellyn who's also planning to upgrade his MkII, so a MkII upgrade seems possible. will keep updated on his progress to see whether it's a good bet to upgrade mine
I'm sure everyone loves hearing theory but somehow I find practical experience with my eyes more trustworthy.
To tell you the truth I am not that bothered by aliasing. In fact it's quite simple to minimize the effect in photoshop with abit of clone stamping.Sure it's not the best thing in the world but frankly it's a tradeoff you'll have to decide if you're willing to live with.
if you've scanned a 4x5 piece of film at high rez before you'll also see aliasing in certain textures.
-
I think "Upgrade" should be in quotes. I'd call it a "downgrade" :-)
As for selective cloning or blurring - yes on mild cases. However, if you get chroma fringes, that'd be a nightmare to remove. And quite frankly, I enjoy taking photos, and I enjoy creative stuff in Photoshop or whatever, but I could not stand for one minute to slave away with my Wacom to remove aliases when a properly designed camera with OLPF minimises or eliminates them. Life's too short..... Is the quest for resolution "that" important. Resolution is part of the image, and quite frankly a small part, well below contrast and below composition and lighting.
Graeme
blahblahblahblah
Let's wait to see more people upgrade their cameras and hopefully post
I've been in touch with Dan Llewellyn who's also planning to upgrade his MkII, so a MkII upgrade seems possible. will keep updated on his progress to see whether it's a good bet to upgrade mine
I'm sure everyone loves hearing theory but somehow I find practical experience with my eyes more trustworthy.
To tell you the truth I am not that bothered by aliasing. In fact it's quite simple to minimize the effect in photoshop with abit of clone stamping.Sure it's not the best thing in the world but frankly it's a tradeoff you'll have to decide if you're willing to live with.
if you've scanned a 4x5 piece of film at high rez before you'll also see aliasing in certain textures.
[a href=\"index.php?act=findpost&pid=147089\"][{POST_SNAPBACK}][/a]
-
I think "Upgrade" should be in quotes. I'd call it a "downgrade" :-)
As for selective cloning or blurring - yes on mild cases. However, if you get chroma fringes, that'd be a nightmare to remove. And quite frankly, I enjoy taking photos, and I enjoy creative stuff in Photoshop or whatever, but I could not stand for one minute to slave away with my Wacom to remove aliases when a properly designed camera with OLPF minimises or eliminates them. Life's too short..... Is the quest for resolution "that" important. Resolution is part of the image, and quite frankly a small part, well below contrast and below composition and lighting.
Graeme
[a href=\"index.php?act=findpost&pid=147094\"][{POST_SNAPBACK}][/a]
Well Graeme I got my composition and lighting and contrast down. so I'm all game for something that can improve the image.
it bugs me when I'm working on my images to see that bloody softness and flatness from my canon files.
I'm sure you understand that the quality of the image can make you feel like working on it all night or just looking at it once and chucking it aside.
-
That's what capture sharpening is for ...
-
Well Graeme I got my composition and lighting and contrast down. so I'm all game for something that can improve the image.
it bugs me when I'm working on my images to see that bloody softness and flatness from my canon files.
I'm sure you understand that the quality of the image can make you feel like working on it all night or just looking at it once and chucking it aside.
[a href=\"index.php?act=findpost&pid=147096\"][{POST_SNAPBACK}][/a]
I'm sure, but that softness is absolutely correct. The alternative is unnatural artifacts and bayer colour fringes on edges. Both, to me, are totally unnaceptable. If it looks soft to you, then sharpen it! And there are sharpening tools now that don't do halos, so you don't have to make it look like a cartoon sketch either.
Or maybe you're staring at the image pixel for pixel and of course it's going to be a tad soft. You're looking at it way too close. You need to get back and take in the entire image. You need to learn to love the OLPF, not fight it. Or, downsample. No other way is completely harmonious with the integrity of the image.
Graeme
-
I'm sure, but that softness is absolutely correct. The alternative is unnatural artifacts and bayer colour fringes on edges. Both, to me, are totally unnaceptable. If it looks soft to you, then sharpen it! And there are sharpening tools now that don't do halos, so you don't have to make it look like a cartoon sketch either.
Or maybe you're staring at the image pixel for pixel and of course it's going to be a tad soft. You're looking at it way too close. You need to get back and take in the entire image. You need to learn to love the OLPF, not fight it. Or, downsample. No other way is completely harmonious with the integrity of the image.
Graeme
[a href=\"index.php?act=findpost&pid=147172\"][{POST_SNAPBACK}][/a]
Graeme, have you ever seen a file from a medium format back enlarged beyond 4ft x 5ft, no AA filter, with bayer colour fringes?
Have you seen a clean AA filtered file enlarged beyond 4ft by 6ft?
Are you saying that if I put an AA filtered image next to a non-AA filtered image, side by side, enlarged to 4 ft, both being equal, I would get an equally detailed image from the AA filtered image than from the non-AA filtered image?
Are you saying there's no loss of detail from an AA filter?
Because from my experience this is not the case, and sharpening past a certain point can make pixels stand out but the detail has turned into clumps already, and the interpolation doesn't create any new detail.
I'm not talking theoretically, I've spent too many hours trying out different methods on different files and ultimately come to the realisation that I prefer a more detail laden file to start with, and moire is an issue I can deal with.
Can we say that all things being equal, and disregarding the moire issue, if I remove the AA filter I am more likely to get an image that is sharper right from the start than an AA filtered one?
And the non-AA filtered image will require less sharpening than the AA filtered image when uprezzed?
How important moire is to each individual is their own issue then, to me it's really not a big problem. I think it's overhyped. I've shot many a textile with medium format backs and the moire issue pops up once in awhile but it's not such a big deal!
-
Sounds like you need to go shoot film and avoid all these issues :-) If you're shooting MF and taking a long time over getting your shot right with regards lighting / exposure and composition, it would be silly for you to spoil your effort by shooting digital if very large prints are what you desire....
Of course there a loss in resolution from the OLPF, but there's an equal loss in aliasing artifacts, moire and aliasing. You can't have one without the other.
But that's the point of having a sensor with an awfully high resolution - it means you can properly anti-alias the incoming light and still have bags of resolution left to spare. If you're shooting 1 or 2 mp, you need all the resolution you can get, and hell at that small count, what's an artifact or three among friends? But as you get beyond that, up past the current mainstream DSLR 10mp or so, and into the realm of 22mp or 39mp, you have a lot of wiggle room, and that means you can properly optically low pass filter your images, downsample them properly and still have a very high resolution image.
I think the problem is that if a camera says it's 10mp, they expect 10mp resolution and they don't get the maths of sampling theorm and understand that to attempt to do so would introduce a vast swathe of unnatural artifacts into the image.
The MF backs probably expect you stop stop way down into diffraction territory so you can stand a hope in hell of getting a decent amount of DOF, and at that point, the diffraction is your OLPF.....
Graeme
-
Sounds like you need to go shoot film and avoid all these issues :-) If you're shooting MF and taking a long time over getting your shot right with regards lighting / exposure and composition, it would be silly for you to spoil your effort by shooting digital if very large prints are what you desire....
Graeme
[a href=\"index.php?act=findpost&pid=147182\"][{POST_SNAPBACK}][/a]
unfortunately I don't have the luxury of shooting film when I am dealing with a group of 20 people and only 5 minutes to get the shot. I have to get it and then finish the shoot in such circumstances, and digital's the only way. Abit of background on what I do so that perhaps you won't think I'm a raving lunatic.
I shoot large groups of people in outdoor environments using strobes. I started out shooting with my 1dsMkII, and then with a Rz67 and a Mamiya 7 once in awhile.
I started shooting 100G on 4x5 film for awhile but that didn't work out due to the lack of speed.
I scan my film on Imacon virtual drum scanners (funny...I prefer the Imacon to drum scans which seem to come out oversharpened and lacking grain usually for some reason)
I print my images up to 5ft for exhibition purposes so we're talking about putting your nose right in front of the print.
I exhibited two images, one shot with a 1DsMkII and one with a H39. now I know it's unfair to compare them since the resolution difference is astounding but one thing I noticed and have been noticing when uprezzing 1DsMkII files is that the interpolation causes edges to have a clumpy (for a lack of better word) look to them, after I sharpen the images.
The H39 file, which never seemed to look that great to me on the computer screen, suddenly popped out in print. everyone noticed it. The details were much finer, edge details especially.
The 1DsMkII file , which I absolutely adore and looks bloody good in my 13x19 portfolio, has a certain flatness to it. When I read on the Hot Rod page about how it looks like your image looks like there's a haze to it and it looks like a photo of a photo rather than a photo alone (does that make sense?), something went off in my head "Bingo!Exactly!"
The 1DsMkII files can look really great with careful processing, but frankly I've always felt that the files lacked a certain depth to them when printed. Editorially I could always tell that my file was a 1DsMkII file whereas it's pretty easy for me to point out when my peers have a film image.
This is just my personal opinion from all the work I've been doing.
Perhaps how the brain and eye perceives and digests visual information and ideas of depth and realism are more minute and diverse than just a simple theoretical explanation.
-
Sounds like you need to go shoot film and avoid all these issues :-) If you're shooting MF and taking a long time over getting your shot right with regards lighting / exposure and composition, it would be silly for you to spoil your effort by shooting digital if very large prints are what you desire....
Of course there a loss in resolution from the OLPF, but there's an equal loss in aliasing artifacts, moire and aliasing. You can't have one without the other.
But that's the point of having a sensor with an awfully high resolution - it means you can properly anti-alias the incoming light and still have bags of resolution left to spare. If you're shooting 1 or 2 mp, you need all the resolution you can get, and hell at that small count, what's an artifact or three among friends? But as you get beyond that, up past the current mainstream DSLR 10mp or so, and into the realm of 22mp or 39mp, you have a lot of wiggle room, and that means you can properly optically low pass filter your images, downsample them properly and still have a very high resolution image.
I think the problem is that if a camera says it's 10mp, they expect 10mp resolution and they don't get the maths of sampling theorm and understand that to attempt to do so would introduce a vast swathe of unnatural artifacts into the image.
The MF backs probably expect you stop stop way down into diffraction territory so you can stand a hope in hell of getting a decent amount of DOF, and at that point, the diffraction is your OLPF.....
Graeme
[a href=\"index.php?act=findpost&pid=147182\"][{POST_SNAPBACK}][/a]
I want to point out also that you'll be really surprised at how different images on screen and images in print turn out. You really can't tell much from a web image, and when you're going into enlargements, that's a whole new ball game. When I'm printing my enlargements I literally have to print strips to compare the contrast between edges to see what I feel is still within the realm of being "photographic", and not cartoonish.
Surprisingly with MF backs I've had situations where I've had to shoot at reaaaaally wide open apertures and they come out great.
This image is on a H39, 35mm F/4
(http://superhyperreal.com/SGIdArm640.jpg)
This image is STUNNING when printed big. in my 13x19 portfolio it's pretty tame. Very strange. Anyway I only noticed abit of moire problems.
This is on a Mamiya 7, 65mm f/4
(http://superhyperreal.com/SGIdgrass640.jpg)
and this is on a Canon 1DsMkII, 35mm f/16
(http://superhyperreal.com/SGIdConst640.jpg)
love love love this image but the problem arose in the small details.
Honestly I do not trust my 1dsMkII for ANY small details. like the faces in the background. They do not enlarge well at all. even when I was using a 16mp Kodak Pro Back the amount of detail you're starting off with when there's no AA filter is just so much finer than with a Canon. This really helps in enlargements.
Regarding bayer colour fringing, I totally agree. In fact when I'm putting my noise up to fashion advert posters I'm looking out for that to guess if the file's shot with a medium format digital back or some other device. Usually I will try to lessen the problem in my files in postproduction.
So I'm willing to live with the issues involved with having no AA filter, and I would like people to know that I think it's not as BAD as everyone seems to think!
btw I've read your recent post in the other forum. Very detailed, learnt abit from there. Thanks. I think the issue I face is the removal of detail by the filter. Would love to see more samples from people who have modded their cameras to confirm what I'm seeing in my experience.
-
How did you shrink those images down for the web - the grass is aliassing very badly in the 2nd jpg (the grass now looks gritty and harsh).
Yes, if you have more detail, you'll be able to uprez better, but any adaptive / intelligent algorithm for uprezzing will get confused by any aliases it sees. Aliases are also bad for "sharpening" as false detail gets exaggerated along with real detail.
Graeme
-
How did you shrink those images down for the web - the grass is aliassing very badly in the 2nd jpg (the grass now looks gritty and harsh).
Yes, if you have more detail, you'll be able to uprez better, but any adaptive / intelligent algorithm for uprezzing will get confused by any aliases it sees. Aliases are also bad for "sharpening" as false detail gets exaggerated along with real detail.
Graeme
[a href=\"index.php?act=findpost&pid=147192\"][{POST_SNAPBACK}][/a]
Hi Graeme, regarding the grass image I use bicubic sharper. I could have made it less gritty and harsh but that is exactly how I want it to look. I exaggerate it almost nearly as much in my other prints too.
I do two things with a file that has the problems of aliasing and bayer color fringing.
It's usually easier to deal with editting out the color fringing problems through careful clonestamping when the image is enlarged actually.
So far the color fringing has not been that big of a deal, occurs on edges which are not heavy with detail in the first place
Aliasing surprisingly has not been much of a problem. Just did a fashion shoot recently where that problem did occur and I clonestamped the hue to make the aliasing even in colour. Left it alone after that because the area affected was minimal.
Not the most optimal of situations I agree, but once again I am aware of the trade off and I find it easy enough to live with, as do alot of other people who deal with their medium format backs that lack AA filters.
Phase One actually mentioned how having more pixels on a sensor helps to reduce moire, which is one of their selling points for upgrading from a P25 (22mp) to P45(39mp)
I'm not sure how true this is, but if this is the case then the 1DsMkII should have less problems with moire using the same logic, even without an AA filter
I think that one has to put into perspective that the number of situations where aliasing comes into effect are not overwhelming. If you're shooting alot of textiles then perhaps that will come into play much more. But I've been shooting people in their clothing for quite awhile now on medium format digital backs and the issue of aliasing pops up so rarely and on such an insignificant scale that I've never felt strongly about it enough to type such tediously long posts on internet forums like this one....
Which leads me to state now that it's friday night here and the club is full of hot chicks waiting for me to pounce on them, so I won't go any further into this. I suggest you try some of the cameras that don't have AA filters to see the benefit, print out some of the images too. I hear the Leica M camera doesn't have one and people seem to like the different feeling the images have.
-
I am in the process of selecting between a Canon 1Ds MK3 and the Nikon D3. The reason for those 2 cameras even being in the same mix is possibly because of this topic.
First, the 21mp appeals to me. But any pictures I can garnish from the internet all have what I call that "Canon Cloud" and I guess i neve realized that it was the AA filter causing it. I just gave that thought over to the possibility that Canon's just never get that "spot on" focus and was something to live with.
On the other hand, almost ALL the samples I am able to find from the Nikon D3 seem to be the opposite. And that caused me to think that the Nikon actually CAN focus spot on. The pictures don't have that cloud and have a clear look to them.
Now the explanation may be different. Maybe the D3 has a considerably weaker AA filter and that would in fact bear out the AA filter argument. So even though the D3 doesn't have the resolution that the sMK3 has, it seems to deliver a "clearer" picture.
Since I also use PK Sharpener, I tried that sharpening that was recomended and it seems to work better than I thought. I tried it on some of my 5D and even a couple of MK3 files and I like the outcome - and not in a small way.
So now the 1Ds MK3 is back on top of my list. I still need to see how much of a handicap it will be in the high ISO compared to what the D3 will deliver because I do a lot more wedding and social work than art or landscape work and low light has a very high place on my list.
Ultimately, I would very much contemplate having the AA filter removed if it became possible to do that. To me, these clinical argument for the AA filter do not go along with real life and remind me very much of my doctor than can only prescrible high priced chemical medicines and balks if i suggest any natural alternatives. For him, there is only one way, but I know that is not true.
So i need to decide between "power files" of the sMK3 or the cleaner higher ISO and clearer pictures of the D3. I lean towards the sMK3 and the luminous sharpening in PK may tip the balance.
Peter
-
Ultimately, I would very much contemplate having the AA filter removed if it became possible to do that. To me, these clinical argument for the AA filter do not go along with real life and remind me very much of my doctor than can only prescrible high priced chemical medicines and balks if i suggest any natural alternatives. For him, there is only one way, but I know that is not true.[a href=\"index.php?act=findpost&pid=147222\"][{POST_SNAPBACK}][/a]
The problem with your analogy is that aliasing is not the natural choice; it is the synthetic one. Things in the real optical world do not snap to grid, subsample, and alias. They just run out of resolution.
I can't explain why so many people like the look of aliasing, and mistake it for subject detail. At the risk of sounding condescending, I have to think that people who like the look of aliasing have a problem with spatial interpretation. Aliasing loses the real center of points of light, of edges, etc, with it's snap-to-grid effect. Everything is choppy and in the wrong place with aliasing, and it is so clearly obvious to my eyes and brain. The images look shattered, in a fixed, repeating pattern.
-
Maxmax.com .
If you look at MF lens MTF curves and think about the MF digital back pixel size, you understand there are not much real resolution there for strong aliasing, or in other words the lens MTF attenuation acts as an AA filter by itself, which is probably a big part that they can go without one in general.
For 35mm DSLR non-landscape works I agree absolutely we need these filters, just as we saw from the Kodak 14n/c examples. 35mm lens can often produce high spacial frequencies over 100lpr/mm while most MF lenses reaches around 60lpr/mm.
The 30% resolution advantage I referred (D2x over 5D) is the amount of true details a careful de-convolution kind of sharpening (definitely not USM) can bring up, such as the Richard-Lucy filter in Raw Developer (the LightRoom came somehow close with the smallest radius but not quite there because of lacking the number of iteration parameter). The R-L deconvolution filter was the one NASA used to fix the fuzzy Hubble Space Telescape images. Again it does not work with all the images but to us "westcoast f64 school" "rock and tree photographers" it is the choice. It is not the edge transition measurements you get from DxO, for example.
LEPING,
Where did you go to have your AA filter removed?
[a href=\"index.php?act=findpost&pid=147077\"][{POST_SNAPBACK}][/a]
-
Even for landscape shots you can need a good AA - grass, in a field as it heads off into the distance will alias quite badly without one.
Graeme
-
Even for landscape shots you can need a good AA - grass, in a field as it heads off into the distance will alias quite badly without one.
[a href=\"index.php?act=findpost&pid=147288\"][{POST_SNAPBACK}][/a]
What's the problem? You don't like blades of grass that come in integer widths, and look like Tetris pieces?
-
FYI attached are 3 in-camera Jpegs with no post processing just size reduction in PS. 5D (no AA filter) 70-200mm2.8 IS. Nothing special just thought you might want to see.
Marc
[attachment=3614:attachment][attachment=3613:attachment][attachment=3615:attachm
ent]
-
FYI attached are 3 in-camera Jpegs with no post processing just size reduction in PS. 5D (no AA filter) 70-200mm2.8 IS. Nothing special just thought you might want to see.
[a href=\"index.php?act=findpost&pid=147304\"][{POST_SNAPBACK}][/a]
Downsized images only distinguish between aliased and non-aliased images by the low-frequency aliasing which may be present, but isn't in this image.
The aliasing obvious in the focused grass area could be obtained *with* an AA filter, and poor downsizing. Do you think that the emphasis on the 45 degree aspects of the grass in the focused area is a good thing? I find it quite annoying, myself.
-
An artifact of bicubic sharper?
Marc
[attachment=3619:attachment]
-
Hi,
There are quite few answers.
1) MF backs do suffer from moire effects
2) Stopping down acts like a AA-filter, because of diffraction
3) High resolution backs have less problems with moire than low resolution backs
4) Lens resolving power does matter, I would suggest that MF-format lenses have less resolving power than premium class 135 lenses. You can check this comparing the "Photodo MTF ratings on:
http://www.photodo.com/products.html?mount...e=Hasselblad+CF (http://www.photodo.com/products.html?mountid=57&name=Hasselblad+CF)
and
http://www.photodo.com/products.html?mount...me=Canon+EF+USM (http://www.photodo.com/products.html?mountid=18&name=Canon+EF+USM)
5) What you photograph does matter
6) How much time you spend on each image in Photoshop matters.
Unfortunately I don't possess an MF digital back, even if I have MF equipment for film, so I don't talk from own experience. Generally speaking you get aliasing when the resolution of the lens exceeds the resolution of the sensor. So what the AA filter does is to reduce resolution. The same can be achieved by stopping down or defocusing. The only situation when you can get problem with moiré is if you are shooting at optimum (medium or large) aperture, have some material having regular structure in the focus plane and using an excellent lens.
Another effect caused by aliasing is that thin branches or twigs on trees can show pixelation or colored pixels. With the APS-C cameras I'm shooting with I have never seen any moire in real life pictures, but they do have AA-filters.
Best regards
Erik
You may want to look at this thread for more discussion on this topic
http://luminous-landscape.com/forum/index....opic=20081&st=0 (http://luminous-landscape.com/forum/index.php?showtopic=20081&st=0)
Would someone like to perhaps enlighten me on why most medium format backs seem to forgo the AA filter?
and medium format backs are held up as the holy grail of image quality
[a href=\"index.php?act=findpost&pid=145765\"][{POST_SNAPBACK}][/a]
-
FYI attached are 3 in-camera Jpegs with no post processing just size reduction in PS. 5D (no AA filter) 70-200mm2.8 IS. Nothing special just thought you might want to see.
Marc
[attachment=3614:attachment][attachment=3613:attachment][attachment=3615:attachm
ent]
[a href=\"index.php?act=findpost&pid=147304\"][{POST_SNAPBACK}][/a]
goddammit stop tempting me!
-
That's a perfect image to show how the aliases fold back to a lower spatial frequency, and hence cannot now be removed by a blur, other than a blur so large that ruins the image and removes all the detail in it. There's no way known to repair an image like that and remove it's aliases.
Graeme
[a href=\"index.php?act=findpost&pid=147088\"][{POST_SNAPBACK}][/a]
I don't believe the image shows specifically how the aliasing folds back to a lower spatial frequency, just that it has in this example.
I presume that for best results, moire effects should be tackled at the RAW image stage if possible.
-
I don't believe the image shows specifically how the aliasing folds back to a lower spatial frequency, just that it has in this example.
I presume that for best results, moire effects should be tackled at the RAW image stage if possible.
[a href=\"index.php?act=findpost&pid=147345\"][{POST_SNAPBACK}][/a]
They are best tackled at the optical state.
-
goddammit stop tempting me!
[a href=\"index.php?act=findpost&pid=147332\"][{POST_SNAPBACK}][/a]
Any camera can give that so-called "3D pop" in a downsized image like that, with the "right" method.
-
I don't think bicubic is a good downsampling filter. Gaussian is "safe" in terms of aliasing, usually, but if you want some visuals on it, http://www.cambridgeincolour.com/tutorials...ize-for-web.htm (http://www.cambridgeincolour.com/tutorials/image-resize-for-web.htm) shows a few examples.
Graeme
An artifact of bicubic sharper?
Marc
[attachment=3619:attachment]
[a href=\"index.php?act=findpost&pid=147320\"][{POST_SNAPBACK}][/a]
-
I don't think bicubic is a good downsampling filter. Gaussian is "safe" in terms of aliasing, usually, but if you want some visuals on it, http://www.cambridgeincolour.com/tutorials...ize-for-web.htm (http://www.cambridgeincolour.com/tutorials/image-resize-for-web.htm) shows a few examples.
Graeme
[a href=\"index.php?act=findpost&pid=147378\"][{POST_SNAPBACK}][/a]
It's sometimes tough to figure out what folks are really trying to say. When I saw the post from marcmccalmont and looked at the attached crop, I naturally figured that he was being a bit tongue in cheek with his "bicubic" comment, since (at least to me) the shoe was a clear moire example in a shot posted to demonstrate how having no filter wasn't a problem (forgive the double negative)
-
Don't read anything into my posts. I'm not smart enough to send hidden messages. The pictures were not intended to be an endorsement to remove your AA filter or not. Just information for those who do not have a camera without a AA filter or experience with one. I believe that for nature and landscapes (random patterns) a AA filter is a hindrance to sharpness. For fashion photography the moiré would be unlivable. For everything in between the camera manufacturers have decided to install one (I think a conservative approach). I personally feel that theory is great but empirical data is more relevant. Practical use has a lot of validity. It is also hard to judge things with low resolution and Jpeg artifacts. The un-compressed 16 bit Tiffs printed on my iPF5000 look much better but I can't show you those. BTW in the crop I don't see the 45 degree moiré in the grass or moiré in Coaches shoes, the pattern is the stitching, the socks look pretty good to me too. What I do like is the 3D pop that the camera has that was not as apparent with the AA filter and the increased sharpness. I would like to see DSLR’s with replaceable AA filters so you can have your cake and eat it too, maybe even IR photography as an option.
Marc
-
I don't think bicubic is a good downsampling filter. Gaussian is "safe" in terms of aliasing, usually, but if you want some visuals on it, http://www.cambridgeincolour.com/tutorials...ize-for-web.htm (http://www.cambridgeincolour.com/tutorials/image-resize-for-web.htm) shows a few examples.
Graeme
[a href=\"index.php?act=findpost&pid=147378\"][{POST_SNAPBACK}][/a]
Interesting link. Is there an easy way to use other algorithms inside of photoshop?
marc
-
An artifact of bicubic sharper?
Marc
Attached Image
I haven't yet seen bicubic-induced color-moire in black socks... So maybe it's something else.
Cheers,
J
-
Hi!
There were couple of interesting threads about mfdbs and moire issues with sharp lenses about half a year ago in LL with very good example shots. For example: http://luminous-landscape.com/forum/index....topic=13563&hl= (http://luminous-landscape.com/forum/index.php?showtopic=13563&hl=)
Cheers,
J
-
The democarcy of these forums is that everyone's opinion is equal. But not all of them are correct. With digital imaging, we are not entiltled to infinite resolution; the attainable resolution is limited by the sampling frequency of the sensor, not the lens. In fact, if you ignore this inconvenient truth, you get aliasing. Aliasing of periodic elements,such as fabrics, are called Moire patterns. The image is distorted, with high frequency details in the wrong location. There is no such thing as color Moire patterns. The sampling problem for Bayer sensors is complicated by the different sampling frequencies of the red, blue and green detector elements. Once this distortion is observed in an image, there is no sure way to remove it unless you know what the image is supposed to look like. In other words there can sometimes be mitigation of the effect but there is no general method to remove aliasing.
The AA fiter is there to attenuate high frequency information beyond that which can be faithfully recorded by the sensor. It is digital imaging 101 to roll off the high frequency information above the Nyquist frequency of the sensor. This is THE standard approach in digital signal processing. No amount of voting or opinion is going to change this. One goes naked at their own peril. In digital photography, the roll off of the high frequency information is done by blurring the image before it reaches the detector with an AA filter. In paractice, AA filters may not be perfect. The goldilocks effect would be not too weak to allow aliasing and not too strong to remove any detail. With the AA filter it is generally helpful to apply a small amount of sharpening during raw conversion.
Arguing, for example, that the Leica M8 is better because it lacks an AA filter, is a hypothesis. There is no proof. It might be some other element in their proprietary development. To prove it is the lack of an AA filter one needs to make measurements with an M8, with and without an AA filter. Measurements, not pictures shot out your porch window that are later displayed as jpegs on a computer monitor.
Sorry fot the rant, but this same subject has just been explored in Cameras, Lenses and Shooting gear.
I don't know what your theories or hypothesis mean. You sound like some informed smart scienticfic dude, but simply...you are wrong. Period. tested and proven. lack of AA delivers sharper, more "3d" images. AA evens things out. Tested and done.
Only problem with the service is that they can only remove 1 of the 2 filters, so it is still not as sharp as MF, or non AA sensors. That is likely why the compared images are not that obviouse of a difference.
-
I don't know what your theories or hypothesis mean. You sound like some informed smart scienticfic dude, but simply...you are wrong. Period. tested and proven. lack of AA delivers sharper, more "3d" images. AA evens things out. Tested and done.
Only problem with the service is that they can only remove 1 of the 2 filters, so it is still not as sharp as MF, or non AA sensors. That is likely why the compared images are not that obviouse of a difference.
Phil,
You seem angry. Why? What proof is there that removing the 2nd AA filter will make the images as sharp as "tested and proven?" You concede that removing one does not provide an impressive result. Showing an image on a web page is not proof. Show me measurements that demonstrate that resolution is improved without artifact.
The interesting thing is that the effect of a well deigned AA filter should diminish as the digital sampling (more mega pixels) increases, until the system is limited by the lens. None of the discussants has observed this effect. Why? Are there other factors that are not considered?
I can imagine building a sensor which is over sampled to the extent that the lens will always be the limiting element. Then the AA filter could be removed and the image could be resampled in the camera and then stored. The highly sampled image will have more noise at each pixel but the resampled image will have the noise level consistent with the lower sampling, just no aliasing.
There is no free lunch. The ultimate resolution of the system is determined by the sampling of the sensor; there is no way to do better. If you remove the AA filter the distortions in the image will be visible when high resolution lenses are used. In fact, these distortions will be present in all images, not just those containing period structures. Details at a spatial frequencies above Nyquist will be rendered at lower frequencies. These distortions cannot be removed by software algorithms.
-
I don't mean to come across as angry. I do see often in forums with people of high degree and regard in the science of optics or theories or mathmetics, ...sometimes negating statements, just because it doesn't make sense on their fact list, etc.
I have used a 1Ds Mk1, and 1DsMk2, and the Kodak SLRc with Leica glass that is sharper than all the Canon L macro lens'. I tested them on a studio stand with the fixed lens on the mount, swapping the bodies..... and the difference is clearly in the image. I don't think I have those exact test samples as it was done for me to know what to work with. I would be willing to do a sample for you.
On that note...Although my Kodak file had a sense of MF quality/razor sharpness, the sensor does not handle differnt light setups well. I would think this is something that has been addressed at some point in the 6 or 7 or so years on sensor dev. So I use a mfdb. I recently used it knowing the lighting would work, and I do have a image from it, but nothing to compare it with.
I don't really know well this relation you explain between sensor and resolving power of the lens = needing a AA filter, but....
Maybe it is time for either Canon or Nikon to buy out the Foveon technology?
-
Hi,
The issue is always complicated by the need of sharpening. An AA-filtered image needs and can take much more sharpening than an unfiltered image.
Regarding the Foevon it would actually just make things worse. It has relatively low resolution so it really needs AA-filtering.
Check this image (from a Sigma DP2)
http://a.img-dpreview.com/reviews/SigmaDP2...res_ACR-003.JPG (http://a.img-dpreview.com/reviews/SigmaDP2/Samples/Comparedto/Res/DP2_res_ACR-003.JPG)
You can observe several inversions of contrast, albeit it's very contrasty. But it shows black lines where they should be white and vice versa.
So the image may be pleasant to the eye but it's mostly artifacts.
The image below is correctly resolved (but much softer)
http://a.img-dpreview.com/reviews/SigmaDP2...III_ACR-003.JPG (http://a.img-dpreview.com/reviews/SigmaDP2/Samples/Comparedto/Res/Ricoh_GRDIII_ACR-003.JPG)
Best regards
Erik
I don't mean to come across as angry. I do see often in forums with people of high degree and regard in the science of optics or theories or mathmetics, ...sometimes negating statements, just because it doesn't make sense on their fact list, etc.
I have used a 1Ds Mk1, and 1DsMk2, and the Kodak SLRc with Leica glass that is sharper than all the Canon L macro lens'. I tested them on a studio stand with the fixed lens on the mount, swapping the bodies..... and the difference is clearly in the image. I don't think I have those exact test samples as it was done for me to know what to work with. I would be willing to do a sample for you.
On that note...Although my Kodak file had a sense of MF quality/razor sharpness, the sensor does not handle differnt light setups well. I would think this is something that has been addressed at some point in the 6 or 7 or so years on sensor dev. So I use a mfdb. I recently used it knowing the lighting would work, and I do have a image from it, but nothing to compare it with.
I don't really know well this relation you explain between sensor and resolving power of the lens = needing a AA filter, but....
Maybe it is time for either Canon or Nikon to buy out the Foveon technology?
-
Hi,
The issue is always complicated by the need of sharpening. An AA-filtered image needs and can take much more sharpening than an unfiltered image.
Regarding the Foevon it would actually just make things worse. It has relatively low resolution so it really needs AA-filtering.
Best regards
Erik
Sharpening after the AA is not resolution gained or even preceived to be gained, specifically on macro work that details are all so important.
I know the res is low. I was hopping one of the big guys can break the res cap?
So anyone have a practical way to explain the reason why we can't have a DSLR without a AA filter...much like the big brothers in MF? I know there is a lens resolution lpmm ratio that I don't know a lot about.... perhaps the spacing between photon sites and the lens resolution spcial thing a ma jig...Like someone said...Its more dangerous to know part of something, than to know nothing of it at all.
-
Hi,
I don't see the point. The AA filter is not as much reducing resolution as reducing contrast. Sharpening increases edge contrast. With macro there are other issues, a macro lens with nominal aperture at f/8 would have a real aperture (with extension taken into account) of f/16, so diffraction can show up early. Once you stop down enough the effect of the AA-filter can be ignored.
The reason that AA-filtering is done is that the sensor cannot reproduce detail that is smaller than a pixel correctly. Lets take one of those power lines polluting the landscape. If the power line is one pixel thick and horizontal it could affect:
1 pixel
2 pixels (if shifted half a pixel)
0 pixels (if it's in between two pixels and the fill factor is low)
So it could vary in thickness, have a staircase effect or simply disappear. Now, the problem doesn't arise very often. Because:
1) Power lines are not normally one pixel wide.
2) Really good technique is needed so that the power line is imaged on a single pixel. Perfect focus, no vibration and medium aperture.
The problem goes away if we reduce pixel size so the pixels are half the size of the smallest spot the lens can reproduce.
The AA-filter is essentially a set of two beamsplitters (one horizontal and one vertical). They split the light beam to make sure that that some of the light also hits the neighboring pixels. The strength of the AA-filter is determined by it's thickness.
I don't know why they don't have AA-filters on MFDBs. One reason is probably that even if artifacting is unavoidable the artifacts are seldom obvious except as colorful moirés. For the problem to show up the feature size should be close to the pixels size.
It seems that Moirés often are a problem when shooting fabric and this is one of the reasons studio photographers want to shoot tethered. They don't want to have 200 shot's with moirés, tethered shooting allows them to discover the problem early.
To explain the problem a little bit better, look at this image taken with a Sigma DP2 (with Foevon sensor without AA-filter)
http://a.img-dpreview.com/reviews/SigmaDP2...res_ACR-002.JPG (http://a.img-dpreview.com/reviews/SigmaDP2/Samples/Comparedto/Res/DP2_res_ACR-002.JPG)
If you look carefully you can see that the lines start to move around at about 14 (half way between 12 and 16) at 18 we have gray and at 20 we see new line structure.
If you now check a DSLR image: http://a.img-dpreview.com/reviews/SigmaDP2...330_ACR-002.JPG (http://a.img-dpreview.com/reviews/SigmaDP2/Samples/Comparedto/Res/Sony_A330_ACR-002.JPG)
You can see that all lines are continous. We se no lines wiggling around and no false contrast.
Best regards
Erik
Sharpening after the AA is not resolution gained or even preceived to be gained, specifically on macro work that details are all so important.
I know the res is low. I was hopping one of the big guys can break the res cap?
So anyone have a practical way to explain the reason why we can't have a DSLR without a AA filter...much like the big brothers in MF? I know there is a lens resolution lpmm ratio that I don't know a lot about.... perhaps the spacing between photon sites and the lens resolution spcial thing a ma jig...Like someone said...Its more dangerous to know part of something, than to know nothing of it at all.
-
Hi,
I don't see the point. The AA filter is not as much reducing resolution as reducing contrast. Sharpening increases edge contrast. With macro there are other issues, a macro lens with nominal aperture at f/8 would have a real aperture (with extension taken into account) of f/16, so diffraction can show up early. Once you stop down enough the effect of the AA-filter can be ignored.
I don't know why they don't have AA-filters on MFDBs.
To explain the problem a little bit better, look at this image taken with a Sigma DP2 (with Foevon sensor without AA-filter)
http://a.img-dpreview.com/reviews/SigmaDP2...res_ACR-002.JPG (http://a.img-dpreview.com/reviews/SigmaDP2/Samples/Comparedto/Res/DP2_res_ACR-002.JPG)
If you look carefully you can see that the lines start to move around at about 14 (half way between 12 and 16) at 18 we have gray and at 20 we see new line structure.
If you now check a DSLR image: http://a.img-dpreview.com/reviews/SigmaDP2...330_ACR-002.JPG (http://a.img-dpreview.com/reviews/SigmaDP2/Samples/Comparedto/Res/Sony_A330_ACR-002.JPG)
You can see that all lines are continous. We se no lines wiggling around and no false contrast.
Best regards
Erik
Thanks for the examples, I had to view my browser at400% to see it.
I do see the lines blurr, but I also see how fuzzy and blurred the overall image is.
If the likelyhood of 1 pixel is low, then why not make it an option? Also, if you are not sure why the MF don't use one, then I don't think you understand it as well as well either.
The edge sharpness is only because there is a difference in pixel color. imagine your image in various areas being smeered this way. It creates an overall fuzzy image and I have done the side by side, and there is a sure difference. Enough for me not to want a filter on it, and take the chance to occasionally get a moire. Why can't they make this a removable filter? perhaps one that screws into the body's lens opening.
-
Hi,
Mamiya had it on the ZD. They had an IR filter as default and an AA filter as option.
I don't know why we have this situation where all DSLRs have AA-filtering and other cameras don't. It seems that all image processing experts on this forum have the view that AA-filtering is needed. Simultaneously, very experienced users of MFDBs state that lack of AA-filtering is an advantage.
Best regards
Erik
Thanks for the examples, I had to view my browser at400% to see it.
I do see the lines blurr, but I also see how fuzzy and blurred the overall image is.
If the likelyhood of 1 pixel is low, then why not make it an option? Also, if you are not sure why the MF don't use one, then I don't think you understand it as well as well either.
The edge sharpness is only because there is a difference in pixel color. imagine your image in various areas being smeered this way. It creates an overall fuzzy image and I have done the side by side, and there is a sure difference. Enough for me not to want a filter on it, and take the chance to occasionally get a moire. Why can't they make this a removable filter? perhaps one that screws into the body's lens opening.
-
I can imagine building a sensor which is over sampled to the extent that the lens will always be the limiting element. Then the AA filter could be removed and the image could be resampled in the camera and then stored. The highly sampled image will have more noise at each pixel but the resampled image will have the noise level consistent with the lower sampling, just no aliasing.
The Fuji sensor attempts to do this by effectively doing a on-sensor bracket-and-blend as well as a few other neat tricks. You get very clean images as a result. Sure, there might be more distortion and noise, but it's very consistent and film-like in that it's hard for our eyes to spot.
BTW,
The Foveon's problem with the resolution tests isn't the aliasing. It's that the software in the camera itself and the processing software as well are trying to obtain "resolution" where the sensor just simply doesn't have any. It's only roughly the same resolution as APS format film - not even 35mm film quality, so they do a lot of software and hardware trickery to smear the blurry areas and make the problem of sub-standard resolution less apparent. Up to the point where it runs out of resolution and they do these "tricks" with the data, it's superbly clean. But ~4MP isn't good enough my anyone's standards. Even the Fuji is barely adequate as a 35mm replacement. (most Dlabs scan at 3000X2000 pixels for 35mm film)
-
The Fuji sensor attempts to do this by effectively doing a on-sensor bracket-and-blend as well as a few other neat tricks. You get very clean images as a result. Sure, there might be more distortion and noise, but it's very consistent and film-like in that it's hard for our eyes to spot.
BTW,
The Foveon's problem with the resolution tests isn't the aliasing. It's that the software in the camera itself and the processing software as well are trying to obtain "resolution" where the sensor just simply doesn't have any. It's only roughly the same resolution as APS format film - not even 35mm film quality, so they do a lot of software and hardware trickery to smear the blurry areas and make the problem of sub-standard resolution less apparent. Up to the point where it runs out of resolution and they do these "tricks" with the data, it's superbly clean. But ~4MP isn't good enough my anyone's standards. Even the Fuji is barely adequate as a 35mm replacement. (most Dlabs scan at 3000X2000 pixels for 35mm film)
Good info! interesting to know the Fuji (S5?) deals with it differently. I would be interested to know how the detail differs.
-
I don't see the point. The AA filter is not as much reducing resolution as reducing contrast. Sharpening increases edge contrast.
I agree with Erik. I'd much rather have the natural, film-like photographs that result from the use of AA filters than the digital-looking images with unnatural artifacts that you get when they're removed.
-
I agree with Erik. I'd much rather have the natural, film-like photographs that result from the use of AA filters than the digital-looking images with unnatural artifacts that you get when they're removed.
I was hoping to get more obective input, although I respect and understand your preference.
I find in my fine art, or street shooting work, I could care less. But when doing a product for a client, I want to see razor sharp. It is not in my financial interest in commercial work, to convince my clients on how technology works, and how the subconsience precieves something is "better". If I have ruled out sharpness as a weak link in my work, I can approach the rest of my shoot day without that being in the back of my mind.
So...why MF has no AA filter, and why does the DSLR's (at least most, not including the Kodak SLRC, 14n) HAVE to have a AA filter?
-
I was hoping to get more obective input, although I respect and understand your preference.
Thanks. The objective part is easy. Removing the AA filter increases aliasing. No ifs, ands, or buts. Just a fact.
The only question is whether aliasing looks good or bad. The answer is necessarily subjective. Some people look at an aliased image and describe it as highly desirable "sharpness" and "3D" look. Others look at the same exact image and see unnatural artifacts.
So...why MF has no AA filter, and why does the DSLR's (at least most, not including the Kodak SLRC, 14n) HAVE to have a AA filter?
I think it's the same reason why MFDB have poor low light performance, low dynamic range for their size, poor sensitivity, slow frame rate, and a host of other deficiencies compared to 35mm DSLRs. Low volume. They have some great advantages too, but unfortunately not enough to stem of death of Medium Format manufacturers (from over a dozen down to just 4 now). They can't afford to do their own sensor design, so they have to buy off-the-shelf parts from Kodak and Dalsa, which have far lower performance (per area) than DSLRS and don't come with AA filters.
When MFDB are used with thin DOF, the aliasing only occurs on the plane of critical focus, such as the eyes and not the hair. When they are used with very deep DOF, diffraction reduces the aliasing.
The higher the resolution, the more likely that motion blur, lens aberrations, diffraction, and other factors will prevent fine detail (and aliasing).
Furthermore, with higher resolution, the less likely that aliasing will contaminate low spatial frequencies (only the higher ones). This means that the aliasing may not be visible in normal sized prints (e.g. magazine spread) even if it is in large ones (40x30).
There is also momentum. Some people have been practicing unsafe imaging (with no protective anti-alias filter) for so long that they have become addicted to the aliasing artifacts. What they see as "sharpness" and "3D", I see as jaggies, stair-stepping, sparkling, "snap to grid", wavy lines, bands, patterns, fringing, popping, strobing, noise, false detail, and of course moiré. You can't rip aliasing away from customers too suddenly, or they will experience extreme withdrawals. Not a good move for a market that is already shrinking.
A good OLPF is lab-grown, high-grade, ground, and polished Lithium Niobate crystal. The cost scales exponentially with area because even the tiniest defect will show up on the image, thanks to being so close to the sensor. Even though MFDB is only three times more area than 35mm, the cost can be an order of magnitude (or more) higher. More importantly, Canon/Nikon ship millions of units a year, compared to less than 6,000/year for all MFDB combined, so economies of scale is a huge factor.
-
Hi,
Just to put things in perspective. When Canon introduced the D1s there was an article here on LL where the author stated that the cost for just the AA-filter alone would be around 2000 USD. *) The AA-filter for the Mamiya ZD did carry a price tag of 2000 USD (or more).
Moiré is a well known problem with MFDBs.
Regarding the artifacting I'd say that we look far to much at images in actual pixels. For real word use the image is always scaled, up or down. When printing the image will be dithered from low PPI to high DPI. We have little control over that process. Sharpening for print is pretty excessive as haloes are introduced to compensate for expected diffusion of ink.
Jonathan Winke, who is writing a lot on this forum, is developing a new image processing tool and is involved in some discussions about down scaling. It seems that aliasing is also problem with downscaling. It seems that the correct way of downscaling an image is to blur it a little bit, downsize and sharpen.
Best regards
Erik
*) http://www.luminous-landscape.com/reviews/...s-Pricing.shtml (http://www.luminous-landscape.com/reviews/cameras/1ds/1Ds-Pricing.shtml)
"Another point to consider is that the 1Ds uses an anti-aliasing filter to reduce moire. I have spoken with some design engineers who tell me that a high quality anti-aliasing filter that size likely costs about $2,000. And, to show that there's logic to this, the optional anti-aliasing filter for the Kodak DCS 760 (a much smaller chip) retails for about $850"
Thanks. The objective part is easy. Removing the AA filter increases aliasing. No ifs, ands, or buts. Just a fact.
The only question is whether aliasing looks good or bad. The answer is necessarily subjective. Some people look at an aliased image and describe it as highly desirable "sharpness" and "3D" look. Others look at the same exact image and see unnatural artifacts.
I think it's the same reason why MFDB have poor low light performance, low dynamic range for their size, poor sensitivity, slow frame rate, and a host of other deficiencies compared to 35mm DSLRs. Low volume. They have some great advantages too, but unfortunately not enough to stem of death of Medium Format manufacturers (from over a dozen down to just 4 now). They can't afford to do their own sensor design, so they have to buy off-the-shelf parts from Kodak and Dalsa, which have far lower performance (per area) than DSLRS and don't come with AA filters.
When MFDB are used with thin DOF, the aliasing only occurs on the plane of critical focus, such as the eyes and not the hair. When they are used with very deep DOF, diffraction reduces the aliasing.
The higher the resolution, the more likely that motion blur, lens aberrations, diffraction, and other factors will prevent fine detail (and aliasing).
Furthermore, with higher resolution, the less likely that aliasing will contaminate low spatial frequencies (only the higher ones). This means that the aliasing may not be visible in normal sized prints (e.g. magazine spread) even if it is in large ones (40x30).
There is also momentum. Some people have been practicing unsafe imaging (with no protective anti-alias filter) for so long that they have become addicted to the aliasing artifacts. What they see as "sharpness" and "3D", I see as jaggies, stair-stepping, sparkling, "snap to grid", wavy lines, bands, patterns, fringing, popping, strobing, noise, false detail, and of course moiré. You can't rip aliasing away from customers too suddenly, or they will experience extreme withdrawals. Not a good move for a market that is already shrinking.
A good OLPF is lab-grown, high-grade, ground, and polished Lithium Niobate crystal. The cost scales exponentially with area because even the tiniest defect will show up on the image, thanks to being so close to the sensor. Even though MFDB is only three times more area than 35mm, the cost can be an order of magnitude (or more) higher. More importantly, Canon/Nikon ship millions of units a year, compared to less than 6,000/year for all MFDB combined, so economies of scale is a huge factor.
-
Neither AA-filtered or unfiltered bayer-sensors work well - they're compromises which became better as higher photosite densities decreased the problem.
MFDBs don't use specialized ASICs which makes them slow and less responsive - that's their only quality-related compromise caused by low production volumes. The rest is technology-related, they set other priorities (just like choosing less "transparent" but more "precise" CFAs - trading sensitivity for color reproduction).
Most DSLR-sensors are also bought by suppliers, as well as AA-filters. Dalsa and Kodak are highly experienced sensor-suppliers manufacturing state-of-the-art-technology even used in much more demanding scientific and military purposes.
Their results have to go through manual post-processing, they're slow and don't offer HD-video, but their per-pixel-IQ is unbeaten - horses for courses.
One of our members made one of the very few useful (same lens!) comparisons regarding sensor/IQ-quality of MFDBs vs. DSLR:
http://luminous-landscape.com/forum/index....showtopic=39318 (http://luminous-landscape.com/forum/index.php?showtopic=39318)
Half as large photosites but comparable microcontrast (despite lower MTF from the lens at the higher frequency) and superior DR.
Lower MTF caused by the lens, AA-filter or sensor/film cannot be compensated by any post-processing.
-
Hi,
MTF is highly sensitive to sharpening at medium frequencies. I highly recommend this article:
http://www.smt.zeiss.com/C12567A8003B8B6F/...Kurven_2_en.pdf (http://www.smt.zeiss.com/C12567A8003B8B6F/EmbedTitelIntern/CLN_31_MTF_en/$File/CLN_MTF_Kurven_2_en.pdf)
The article refers to a couple of images illustrating different MTFs, there are some companion pictures like this:
http://www.zeiss.de/C12567A8003B8B6F/Graph...le/Image_02.jpg (http://www.zeiss.de/C12567A8003B8B6F/GraphikTitelIntern/CLN31MTF-KurvenBild2/$File/Image_02.jpg)
I have printed the image in A2, and read the article. It's an interesting experience.
Best regards
Erik
Neither AA-filtered or unfiltered bayer-sensors work well - they're compromises which became better as higher photosite densities decreased the problem.
MFDBs don't use specialized ASICs which makes them slow and less responsive - that's their only quality-related compromise caused by low production volumes. The rest is technology-related, they set other priorities (just like choosing less "transparent" but more "precise" CFAs - trading sensitivity for color reproduction).
Most DSLR-sensors are also bought by suppliers, as well as AA-filters. Dalsa and Kodak are highly experienced sensor-suppliers manufacturing state-of-the-art-technology even used in much more demanding scientific and military purposes.
Their results have to go through manual post-processing, they're slow and don't offer HD-video, but their per-pixel-IQ is unbeaten - horses for courses.
One of our members made one of the very few useful (same lens!) comparisons regarding sensor/IQ-quality of MFDBs vs. DSLR:
http://luminous-landscape.com/forum/index....showtopic=39318 (http://luminous-landscape.com/forum/index.php?showtopic=39318)
Half as large photosites but comparable microcontrast (despite lower MTF from the lens at the higher frequency) and superior DR.
Lower MTF caused by the lens, AA-filter or sensor/film cannot be compensated by any post-processing.
-
Daniel or Erik...can you post an image as to what you are referring about? (please keep them at 100% for obviouse reasons, as I don't care to look into something beyond its ability to print or reproduce
I have been shooting with MFdB for a few years now, and have not seen or noticed the artifacts you mention.
meanwhile, thanks for these links all, I will get busy reading them.
-
I have started to read the links, and after looking at the twin lens images of Zeiss post..and when I got to the few paragraphs in the intro...Are todays lens good enough for 24mp sensors....
I thought...Why don't I just conduct a test and be done with it...As the other link where Chris posted his P65 to the Niko D3. Its not a convincing test for detail res.
In my opinion, As much as I read, respect and enjoy seeing his work, I think the test is very limited for seeing detail and this limits the application of it for this discussion. I can clearly see its value as I think his intention was more so for DR, exposure highlight and such...(at least I see it more applicable for that). But if you look at the image and the single light source, the image is rather compressed of detail. there is no revealing light source, only a reflective backlighting that highlights and kisses the subjects. so contrast is very high, color saturation is low. The yarn balls with fuzz on the table is a good area to look at, but I am not convinced as it lacks much info for such a scrutany.
So...I will conduct a test. I will use the same lens, on the same stand, fixed, and only swap the camera bodies. I will use a subject with detail, and light it with 2 or 3 fixed heads. I will stop the lens down to 8, 11 and 22 as I noticed the Leica I have can do 22 before the light starts edging on the blades and defracts.. Does anyone want to ask or suggest anything more before I do this test, or confirm some obvious bits of info that might be overlooked?
Even a subject matter? I don't want to do the test and then have something I left out.
I will look over and post the images. Maybe you eagle eyes of jags and artifacts can point things out.
btw, have any of you done this already?....feel free to post.
-
Daniel or Erik...can you post an image as to what you are referring about?
Sure thing. The differences are sometimes very subtle. It's a good idea to first examine an image where the aliasing is very obvious and pronounced. Here is one such demonstration: the first image has lots of aliasing, the second has very little:
(http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-point.png) (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-point.png)
(http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-lanczos.png) (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-lanczos.png)
I suggest opening them in browser tabs for A/B comparison. Here are a few examples of what is different:
- Speckling in the hair (not there in real life).
- Individual hairs of stubble (they should be too small to see).
- Ear is jagged.
- Catchlight zigzags
Those are examples of the kinds of artifacts that make aliased images look fake and unnatural to me.
-
Hi,
I would be very, very, very surprised if you'd need to stop down to f/22 for diffraction. Essentially all lenses start to show diffraction past f/8. Your lenses should probably reach maximum performance at f/4.
The measurement I show below was taken on a Minolta 80-200/2.8 APO zoom, it's quite obvious that maximum performance is reached around f/8. The factor that limits performance on lenses is diffraction. Essentially all aberrations except lateral chroma and diffraction increase with aperture.
http://83.177.178.7/ekr/images/stories/difractionlimit.gif (http://83.177.178.7/ekr/images/stories/difractionlimit.gif)
Admittedly, the loss of sharpness to diffraction is not easy to see, but as you can see in the figure above, you are loosing half of the resolution at f/22. This essentially means that we turned that 10 MPixel APS-C camera in a 2.5 MPixel APS-C camera.
A major factor is of course focusing. Achieving correct focus can be very hard. Lloyd Chambers has reported some issues about focusing the Leica M9 on his pay site. Erwin Puts considers a viewfinder loupe essential on the M9 for correctly focusing longer lenses. On DSLRs the best way of achieving correct focus is "live view".
And don't forget AA-filtered images need significant sharpening, very high amount at very low radius.
In my view flowers are pretty good natural test subjects. Using a Dollar bill is also quite nice,
like this one: http://www.luminous-landscape.com/images-b...est%20Image.jpg (http://www.luminous-landscape.com/images-bill/Full%20Test%20Image.jpg)
from this test: http://www.luminous-landscape.com/essays/back-testing.shtml (http://www.luminous-landscape.com/essays/back-testing.shtml)
I'd shoot the Dollar bill in half the scale because I feel it lack fine enough detail in the original tests. The Dollar bill is nice because many of us happen to have some around.
I would also try some focus variations. I'm using an RRS MPR rail to achieve that, keeping focus and shifting camera back an forth.
Check here: http://www.slrgear.com/articles/focus/focus.htm (http://www.slrgear.com/articles/focus/focus.htm)
and perhaps also this:
http://www.imx.nl/photo/leica/camera/page155/m9part2.html (http://www.imx.nl/photo/leica/camera/page155/m9part2.html)
All that said I'm looking forward to your findings!
Best regards
Erik
I have started to read the links, and after looking at the twin lens images of Zeiss post..and when I got to the few paragraphs in the intro...Are todays lens good enough for 24mp sensors....
I thought...Why don't I just conduct a test and be done with it...As the other link where Chris posted his P65 to the Niko D3. Its not a convincing test for detail res.
In my opinion, As much as I read, respect and enjoy seeing his work, I think the test is very limited for seeing detail and this limits the application of it for this discussion. I can clearly see its value as I think his intention was more so for DR, exposure highlight and such...(at least I see it more applicable for that). But if you look at the image and the single light source, the image is rather compressed of detail. there is no revealing light source, only a reflective backlighting that highlights and kisses the subjects. so contrast is very high, color saturation is low. The yarn balls with fuzz on the table is a good area to look at, but I am not convinced as it lacks much info for such a scrutany.
So...I will conduct a test. I will use the same lens, on the same stand, fixed, and only swap the camera bodies. I will use a subject with detail, and light it with 2 or 3 fixed heads. I will stop the lens down to 8, 11 and 22 as I noticed the Leica I have can do 22 before the light starts edging on the blades and defracts.. Does anyone want to ask or suggest anything more before I do this test, or confirm some obvious bits of info that might be overlooked?
Even a subject matter? I don't want to do the test and then have something I left out.
I will look over and post the images. Maybe you eagle eyes of jags and artifacts can point things out.
btw, have any of you done this already?....feel free to post.
-
Good info! interesting to know the Fuji (S5?) deals with it differently. I would be interested to know how the detail differs.
The F5 was the old technology. The new uses a similar technology but they also have arranged the (sub)pixels in a diagonal grid and also do some fancy binning as well before the bracketing and blending is applied.
Because our eyes don't track diagonal displacement as easily and they can arrange the sensor in effectively a solid stacked array(as opposed to a triangle type formation in a Bayer sensor) - just on a slightly tighter diagonal. A typical Bayer sensor suffers a roughly .6-.7 efficiency ratio in each dimension compared to film's real pixels at each location. The Fuji, with the blending applied, though, it creates a .9 or better ratio. It's close but roughly 1MP short of 35mm scanned film. Still, to do the same with a Bayer sensor, you need a whopping ~16MP due to the efficiency losses. And that creates a lot of artifacts and errors in the image that must either be downsampled or adjusted with various hardware and software trickery. The software is great at what it does lately, but it's really fancy kludging. Bayer sensors have issues that can't be worked around past what we currently see with the D3x and similar cameras. Something new is required at this point to make any significant gain in image quality.
http://www.fujifilm.com/products/digital_c...s5pro/features/ (http://www.fujifilm.com/products/digital_cameras/s/finepix_s5pro/features/)
This is the older B&B technology and it works very very well.
http://www.fujifilm.com/products/digital_c...es/page_02.html (http://www.fujifilm.com/products/digital_cameras/s/finepix_s200exr/features/page_02.html)
This is the new technology which they unfortunately aren't putting in a DSLR body or making higher resolution. The three modes are high resolution which is standard DSLR and useless, the HDR mode, which is also no better than other sensors, and the HQ mode which is what should be used nearly all the time as default - because it gives you the clean look that you want. But you do have to apparently defeat the on-board AI that wants to select what it thinks is best and just use this mode.
Huge pain. I hate it when electronics try to out-think you.
Of course, 6MP in HQ mode also limits the resolution of fine details too greatly. That's the real issue with the Foveon and Fuji sensors. Unlike the D3x which is approaching the limits of the optics in fine details, they are only halfway there. If that.
It's so close, but not in a proper DSLR type package(WHY???) and about half the MP it needs to be(12MP with this sensor would flat out punk everything short of a DB, yet Fuji doesn't seem to want to go there for some odd reason) And Sigma has done zilch with their sensor as well.
-
Hi,
I have not done any testing on AA-filter less cameras as I don't have any.
I have done some testing 6x7 film against digital. Whatever your results are someone is sure to complain. In my view the kind of testing to do can be valuable to:
- Potential M9 buyers
- Any one considering his camera for AA-filter removal
It may also help to increase our understanding of the issues at hand.
But, someone will always find some error in your testing. Be prepared for some "flak".
Best regards
Erik
I have started to read the links, and after looking at the twin lens images of Zeiss post..and when I got to the few paragraphs in the intro...Are todays lens good enough for 24mp sensors....
I thought...Why don't I just conduct a test and be done with it...As the other link where Chris posted his P65 to the Niko D3. Its not a convincing test for detail res.
In my opinion, As much as I read, respect and enjoy seeing his work, I think the test is very limited for seeing detail and this limits the application of it for this discussion. I can clearly see its value as I think his intention was more so for DR, exposure highlight and such...(at least I see it more applicable for that). But if you look at the image and the single light source, the image is rather compressed of detail. there is no revealing light source, only a reflective backlighting that highlights and kisses the subjects. so contrast is very high, color saturation is low. The yarn balls with fuzz on the table is a good area to look at, but I am not convinced as it lacks much info for such a scrutany.
So...I will conduct a test. I will use the same lens, on the same stand, fixed, and only swap the camera bodies. I will use a subject with detail, and light it with 2 or 3 fixed heads. I will stop the lens down to 8, 11 and 22 as I noticed the Leica I have can do 22 before the light starts edging on the blades and defracts.. Does anyone want to ask or suggest anything more before I do this test, or confirm some obvious bits of info that might be overlooked?
Even a subject matter? I don't want to do the test and then have something I left out.
I will look over and post the images. Maybe you eagle eyes of jags and artifacts can point things out.
btw, have any of you done this already?....feel free to post.
-
Sure thing. The differences are sometimes very subtle. It's a good idea to first examine an image where the aliasing is very obvious and pronounced. Here is one such demonstration: the first image has lots of aliasing, the second has very little:
(http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-point.png) (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-point.png)
(http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-lanczos.png) (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-lanczos.png)
I suggest opening them in browser tabs for A/B comparison. Here are a few examples of what is different:
- Speckling in the hair (not there in real life).
- Individual hairs of stubble (they should be too small to see).
- Ear is jagged.
- Catchlight zigzags
Those are examples of the kinds of artifacts that make aliased images look fake and unnatural to me.
I see exactly what you see here, and that is horrible.
I have not had 1 image with this artifact in my years of using a AA free MFdB. Nor have I seen this in using my Kodak SLRc...which I think has issues in areas with color. I am regularly opening these files in 400 or more % as I retouch my own work, and this is NEVER been something I have seen. This looks like JPEG down compression artifacts.
Maybe you are referring to some other format besides RAW? Or perhaps there is some lens to sensor relation, but I doubt it, as I have used 16 to 400mm of just about anything in between.
-
Hi,
I would be very, very, very surprised if you'd need to stop down to f/22 for diffraction. Essentially all lenses start to show diffraction past f/8. Your lenses should probably reach maximum performance at f/4.
The measurement I show below was taken on a Minolta 80-200/2.8 APO zoom, it's quite obvious that maximum performance is reached around f/8. The factor that limits performance on lenses is diffraction. Essentially all aberrations except lateral chroma and diffraction increase with aperture.
http://83.177.178.7/ekr/images/stories/difractionlimit.gif (http://83.177.178.7/ekr/images/stories/difractionlimit.gif)
Admittedly, the loss of sharpness to diffraction is not easy to see, but as you can see in the figure above, you are loosing half of the resolution at f/22. This essentially means that we turned that 10 MPixel APS-C camera in a 2.5 MPixel APS-C camera.
A major factor is of course focusing. Achieving correct focus can be very hard. Lloyd Chambers has reported some issues about focusing the Leica M9 on his pay site. Erwin Puts considers a viewfinder loupe essential on the M9 for correctly focusing longer lenses. On DSLRs the best way of achieving correct focus is "live view".
And don't forget AA-filtered images need significant sharpening, very high amount at very low radius.
In my view flowers are pretty good natural test subjects. Using a Dollar bill is also quite nice,
like this one: http://www.luminous-landscape.com/images-b...est%20Image.jpg (http://www.luminous-landscape.com/images-bill/Full%20Test%20Image.jpg)
from this test: http://www.luminous-landscape.com/essays/back-testing.shtml (http://www.luminous-landscape.com/essays/back-testing.shtml)
I'd shoot the Dollar bill in half the scale because I feel it lack fine enough detail in the original tests. The Dollar bill is nice because many of us happen to have some around.
I would also try some focus variations. I'm using an RRS MPR rail to achieve that, keeping focus and shifting camera back an forth.
Check here: http://www.slrgear.com/articles/focus/focus.htm (http://www.slrgear.com/articles/focus/focus.htm)
and perhaps also this:
http://www.imx.nl/photo/leica/camera/page155/m9part2.html (http://www.imx.nl/photo/leica/camera/page155/m9part2.html)
All that said I'm looking forward to your findings!
Best regards
Erik
Hi Erik,
I will properly focus both using the same detail area as referrence.
I think different lens combo will have difference deffraction points. But I will do it with your F4, as well as a couple other F stops.
I will use the lens mount on to the Foba studio stand that doesnt budge. the only moving part will be the body.
Money is good, glass, metals, and some fuzzy subject. I am not sure about flowers as they are organic and perhaps if I get close enough to see the veiny cell seperatation...but I think I can manage a leaf.branch without going to the florist.
***I want to evaluate the images at there RAW sharpness, but I am willing to sharpen them equally, as it defeats the purpose of the test to sharpen one file more than the other.
My point is to get the best file to start work from...Not to massage the best file.
-
The F5 was the old technology. The new uses a similar technology but they also have arranged the (sub)pixels in a diagonal grid and also do some fancy binning as well before the bracketing and blending is applied.
Because our eyes don't track diagonal displacement as easily and they can arrange the sensor in effectively a solid stacked array(as opposed to a triangle type formation in a Bayer sensor) - just on a slightly tighter diagonal. A typical Bayer sensor suffers a roughly .6-.7 efficiency ratio in each dimension compared to film's real pixels at each location. The Fuji, with the blending applied, though, it creates a .9 or better ratio. It's close but roughly 1MP short of 35mm scanned film. Still, to do the same with a Bayer sensor, you need a whopping ~16MP due to the efficiency losses. And that creates a lot of artifacts and errors in the image that must either be downsampled or adjusted with various hardware and software trickery. The software is great at what it does lately, but it's really fancy kludging. Bayer sensors have issues that can't be worked around past what we currently see with the D3x and similar cameras. Something new is required at this point to make any significant gain in image quality.
http://www.fujifilm.com/products/digital_c...s5pro/features/ (http://www.fujifilm.com/products/digital_cameras/s/finepix_s5pro/features/)
This is the older B&B technology and it works very very well.
http://www.fujifilm.com/products/digital_c...es/page_02.html (http://www.fujifilm.com/products/digital_cameras/s/finepix_s200exr/features/page_02.html)
This is the new technology which they unfortunately aren't putting in a DSLR body or making higher resolution. The three modes are high resolution which is standard DSLR and useless, the HDR mode, which is also no better than other sensors, and the HQ mode which is what should be used nearly all the time as default - because it gives you the clean look that you want. But you do have to apparently defeat the on-board AI that wants to select what it thinks is best and just use this mode.
Huge pain. I hate it when electronics try to out-think you.
Of course, 6MP in HQ mode also limits the resolution of fine details too greatly. That's the real issue with the Foveon and Fuji sensors. Unlike the D3x which is approaching the limits of the optics in fine details, they are only halfway there. If that.
It's so close, but not in a proper DSLR type package(WHY???) and about half the MP it needs to be(12MP with this sensor would flat out punk everything short of a DB, yet Fuji doesn't seem to want to go there for some odd reason) And Sigma has done zilch with their sensor as well.
Wish they would further dev these technologies into recent sizes
I want to take the images, and if it is hard for me to find a difference, I will not think of it any more, unless I get a new body to see if the AA has changed strength, and to what degree.
If I see the difference and it is fairly easy to see...I will still be in a wonder why there is no OPTIONAL AA, or versions without AA.
As I stated earlier, and the ports that Daniel posted...
I have never seen these artifacts in my RAW files using DSLR or MFdB that have no AA.
-
Hi,
I have not done any testing on AA-filter less cameras as I don't have any.
I have done some testing 6x7 film against digital. Whatever your results are someone is sure to complain. In my view the kind of testing to do can be valuable to:
- Potential M9 buyers
- Any one considering his camera for AA-filter removal
It may also help to increase our understanding of the issues at hand.
But, someone will always find some error in your testing. Be prepared for some "flak".
Best regards
Erik
Thanks for the heads up, I have seen the flak on others..:-)
I cannot see how you can do apple to apple when you are using a differnt "FILM" If we were testing Velvia 50 vs Velvia 50 with some brand filter vs no brand filter and we were getting softer images...OK,
but doing a Film to Sensor. That is another test that I am not interested in. I stopped processing E6 and going to labs a few years back :-)
I was hoping to get manufacturers listening so they can give us an OPTION....IF the test finds that there is enough softness to desire more out of AA DSLR's.
As I mention, I had done this test, and I did find it to be sharper using the KODAK SLRC(13+mp) body vs the 1ds(11+mp). But because of all the color issues, I used the 1Ds more, but shortly after I got the MFdB.
-
I see exactly what you see here, and that is horrible.
Excellent! Most people who see this picture prefer the aliased version, so it's nice to meet another person who dislikes the aliasing in it as much as I do, at least for images where the aliasing is extremely bad.
Now the only question is about circumstances where the aliasing is much more subtle (which is usually the case when comparing filtered-vs-filterless cameras).
Maybe you are referring to some other format besides RAW?
I am talking about raw. The reason I gave you the extreme example was to set a baseline. If you had preferred the aliased image in the extreme example (as many do), then I would know that you prefer the aliased image in a more subtle example for sure.
Now let's take a look at the following somewhat flawed comparison (http://forums.dpreview.com/forums/read.asp?forum=1019&message=30112382) between an SD14 (no AA filter) and a 50D (relatively weak AA filter).
Here is the SD14 100% crop:
(http://thebrownings.name/photo/2010-01-aliasing/106242475.k6guNjIC.SDIM3020-crop.png)
And here is the 50D 100% crop (after downsampling to the same pixel count):
(http://thebrownings.name/photo/2010-01-aliasing/IMG_3026-rawtherapee-4mp-crop.png)
The full SD14 image (http://thebrownings.name/photo/2010-01-aliasing/106242475.k6guNjIC.SDIM3020.jpg)
The full 50D image. (http://thebrownings.name/photo/2010-01-aliasing/IMG_3026-rawtherapee-4mp.png)
The purpose of the comparison was not to show the difference between filtered and unfiltered cameras, so it is somewhat flawed for the purpose of this thread, but I think it can still illustrate some of the types of aliasing artifacts that are seen with filterless cameras.
This example does not have aliasing that is quite as extreme as my last example, but it is a man-made subject, where more people tend to object to the aliasing artifacts.
For example, there appear to be many missing rivets in the SD14 image, while they are all there in the 50D image.
Do you feel the same way about this aliased image as you did the last (i.e. "horrible")? Or do you see the aliased image as having better sharpness/3D than the anti-aliased one? If the former, then the next task will be to find a comparison where the difference in aliasing is more subtle.
-
If I see the difference and it is fairly easy to see...
I'm sure you will see the difference. The AA filter reduces contrast at spatial frequencies 30% lower than needed to suppress aliasing, so the downside is very easy to see. The only question is whether the loss in contrast is worth the decrease in artifacts. For me, the answer is a resounding yes, and I feel that most of the contrast can be recovered with sharpening. Others see no difference in artifacts, and/or disagree about the sharpening, so to them the loss in contrast has no upside.
I will still be in a wonder why there is no OPTIONAL AA, or versions without AA.
My guess is that it's because the market is not large enough to support it. It would have to be an all new DSLR line, like the difference between Canon 1000D and 400D. The new filterless line of DSLR would need to have as many sales as the other lines, and manufacturers must feel that it will not, otherwise I'm sure they would do it for the money. Or perhaps it's because the support expenses (complaints/returns due to artifacts) outweigh the increased sales profits.
-
Now let's take a look at the following somewhat flawed comparison (http://forums.dpreview.com/forums/read.asp?forum=1019&message=30112382) between an SD14 (no AA filter) and a 50D (relatively weak AA filter).
The purpose of the comparison was not to show the difference between filtered and unfiltered cameras, so it is somewhat flawed for the purpose of this thread, but I think it can still illustrate some of the types of aliasing artifacts that are seen with filterless cameras.
This example does not have aliasing that is quite as extreme as my last example, but it is a man-made subject, where more people tend to object to the aliasing artifacts.
For example, there appear to be many missing rivets in the SD14 image, while they are all there in the 50D image.
Do you feel the same way about this aliased image as you did the last (i.e. "horrible")? Or do you see the aliased image as having better sharpness/3D than the anti-aliased one? If the former, then the next task will be to find a comparison where the difference in aliasing is more subtle.
This is not an really good comparison.
I made these observations below from the jpegs...BUT I will ignore them. they mean little to me. I don't have a RAW vs RAW. The download of the raw is not available for either file (But I did download the other files in raw and will check later)
50D is known to have some issues as the microlens have no gapping. The SD14 is upresed to the 50D size. (yet still looks really good in comparison).
these are OUT of CAMERA JPEGS.
Oddly enough on this sample the SD14 is more attractive to me... Yes you are right the subject makes it so...machined subjects have more pop. (50D has a magenta cast that is bothersome and can muddle things also)
Did you mean...People prefer the unfiltered image on machine made subjects as they are straight lines and clear seperations vs organic ?
I am going to guess on this and may be obviouse....When a AA fliter is used, the sensor is less sensitive to be "judgmental" of assigning an 0 or a 1, or any number to identify a hue. Without it, it makes a more "cut" decision.
For me, it looks more clear cut. as I see the rivets on the SD14 may be blurred to the color on a few, but on the 50D they are all blurred.
-
BTW....I am not finding the artifacts in these 2 samples....I do notice a few rivets blurred, and I will look at it more closely and other areas...but it is NO point doing it on a JPEG.
-
I don't have a RAW vs RAW. The download of the raw is not available for either file
Both raw files can be downloaded from the link in my post. Here are direct links:
SDIM3020.X3F (http://thebrownings.name/photo/2010-01-aliasing/SDIM3020.X3F)
IMG_3026.CR2 (http://thebrownings.name/photo/2010-01-aliasing/IMG_3026.CR2)
The SD14 is upresed to the 50D size. (yet still looks really good in comparison). [...] Oddly enough on this sample the SD14 is more attractive to me.
OK, it appears this is where our perceptions begin to differ. I find the SD14 to be inferior in every way, and especially so when upsampled.
Yes you are right the subject makes it so...machined subjects have more pop. Did you mean...People prefer the unfiltered image on machine made subjects as they are straight lines and clear seperations vs organic ?
Actually, I meant the opposite. The preference for aliasing can be split along these lines:
- Always likes aliasing
- Likes aliasing on organic subjects but not man-made
- Likes aliasing on man-made but not organic subjects
- Always dislikes aliasing
My experience is that the most common preference is "Likes aliasing on organic subjects but not man-made".
I am going to guess on this and may be obviouse....When a AA fliter is used, the sensor is less sensitive to be "judgmental" of assigning an 0 or a 1, or any number to identify a hue. Without it, it makes a more "cut" decision.
I think that's a good analogy.
as I see the rivets on the SD14 may be blurred to the color on a few, but on the 50D they are all blurred.
This is a significant departure in our tastes. To me, it is much better to have an image that tells you "there are a bunch of small rivets here, but they're blurry and hard to see" rather than "here are a bunch of sharp and detailed rivets, and by the way, some of them are completely missing." I'd rather have a blurry and realistic view of something than a sharp and false view.
-
Both raw files can be downloaded from the link in my post. Here are direct links:
SDIM3020.X3F (http://thebrownings.name/photo/2010-01-aliasing/SDIM3020.X3F)
IMG_3026.CR2 (http://thebrownings.name/photo/2010-01-aliasing/IMG_3026.CR2)
OK, it appears this is where our perceptions begin to differ. I find the SD14 to be inferior in every way, and especially so when upsampled.
Actually, I meant the opposite. The preference for aliasing can be split along these lines:
- Always likes aliasing
- Likes aliasing on organic subjects but not man-made
- Likes aliasing on man-made but not organic subjects
- Always dislikes aliasing
My experience is that the most common preference is "Likes aliasing on organic subjects but not man-made".
I think that's a good analogy.
This is a significant departure in our tastes. To me, it is much better to have an image that tells you "there are a bunch of small rivets here, but they're blurry and hard to see" rather than "here are a bunch of sharp and detailed rivets, and by the way, some of them are completely missing." I'd rather have a blurry and realistic view of something than a sharp and false view.
Man made?, but not organic? thats a bit confusing for me. that structure is assembled by man...those parts are NOT man made.
I will post more feedback when I get back :-)
-
Man made?, but not organic? thats a bit confusing for me. that structure is assembled by man...those parts are NOT man made.
Technically speaking, the portrait I posted above is also man-made (though I consider it "organic").
-
Oddly enough on this sample the SD14 is more attractive to me...
That's the difference, though. The SD14 looks smoother and has more accurate colors overall. Though it obviously is suffering from resolution and moire issues, they are all blurred the same wherever they happen. It really looks like you're shooting with poor definition film and a lens with less than perfect optics as opposed to digital. At least in my opinion. But that's still a very pleasing result if you're taking scenery and other shots where every detail isn't critical.
Now, a lot of people probably don't like that "look", and that's fine as well. But Bayer sensors also have their share of problems with color balance and obvious blurring of certain parts of the image as they try to compensate. You'll often see red areas on roses for instance that are very blurry and ill-defined and green areas that pop out at you and are crisp in the same image.
http://www.ddisoftware.com/sd14-5d/ (http://www.ddisoftware.com/sd14-5d/)
A good writeup of the differences. Is one better than the other? Hard to say either way, since both have areas that they excel at and areas where you wish you could kick it in the rear end to make it work properly.
But a few highlights:
- The Foveon sensor has 1700 LPI resolution no matter what it's shooting at. Bayer sensors as you can see in the chart 1/4 the way down are all over the place for various colors(and why black and white resolution tests are junk - they must be color. Purple is a known problem area for Bayer sensors. And I'd rather it blow out yellowish like film and how our eyes see sunlight affecting objects than pure white.
- 2/3 the way down it shows how this can affect color images. I'd personally rather look at consistent results than varying resolutions, but then again, the Foveon/Sigma sensor is just too low. The sensor is effectively APS-C size and resolution(bit smaller actually). Huge fail here. But the results do look very nice for what it does render.
- The bottom images also show it well. The Bayer sensor has fairly cold and accurate images, but that's not what we see with our eyes(the SD14 is off from a technical perspective but is correct from a visual one)
http://www.polas.net/sigma/sd14/d200/sd14_vs_d200.php (http://www.polas.net/sigma/sd14/d200/sd14_vs_d200.php)
Here it's a bit more mixed. Sigma wins for color and highlights in my book, but has a really awful DR because the sensor effectively is ISO 50 film and the rest is digital trickery and boosting of the signals. That's perhaps the biggest failing of the technology in that it's too much like film in this respect. The Fuji, otoh, has no such issues. The high quality mode does great with DR as well. And you can recover most photos that are slightly washed out in processing, which is impossible with other sensors.
note - his claims, IMO, are wrong about the highlights, as sunlight is *yellow* and our eyes expect yellowish white where there's too much sun and not pure white. True they do blow out far sooner, but it's color-correct over-saturation. It looks like you took Velvia and put it in a half-frame camera or old 35mm pocket camera. Of course, one advantage of both the Fuji and Sigma bodies is that they accept Nikon lenses(good choice here) and any crappy lens will produce the same consistent results(sigh). This can mean a good price savings but as a person who has dealt with medium format and black and white photography as well, it's too much to swallow. In fact, I had a SD10 a few years back and sold it because it was giving me beautiful but pocket camera results that were virtually identical to my old 35mm Konica rangefinder (circa 1980 or so - still sees use on trips as a backup). I'd hoped that the SD15 was a full frame sensor but no luck. And Fuji's idiocy boggles my mind. Best sensor on the market in a non SLR body...
-
Technically speaking, the portrait I posted above is also man-made (though I consider it "organic").
I would 100% agree that it is organic, but God might interject on it being man made. :-P
well, I setup most of my subject material, and mounted the lens, . I have a couple bodies and lens' getting serviced, and should be back any day. So I think by Monday or so I should have taken the pictures.
I still don't know what you mean by artifact, as I have still not seen any sample other than the first portrait which again, I have not seen this in the years of shooting I have done. And on the SD14 vs 50D....2 bad examples. Sorry.
Plekto, Interesting sensor comparison...I will read the link when i can. But So far thats you and I preferring the SD14 image :-)
I don't know where Daniel gets "people prefer the blurred look" data, and I honestly can not give it any value. And I am not sure how you have quantified your experience? How many people have you done this side by side study comparison...and why? Why not see what you like, or perhaps you have done it on behalf of sensor mfg-ers?
I don't want to sound odd about it...but the claims that most people prefer this over the other claim IS a bit odd to me.
Regardless, this is good posting, and hope I can set aside this ponder and spend more time shooting.
-
Excellent! Most people who see this picture prefer the aliased version, so it's nice to meet another person who dislikes the aliasing in it as much as I do, at least for images where the aliasing is extremely bad.
Now the only question is about circumstances where the aliasing is much more subtle (which is usually the case when comparing filtered-vs-filterless cameras).
I am talking about raw. The reason I gave you the extreme example was to set a baseline. If you had preferred the aliased image in the extreme example (as many do), then I would know that you prefer the aliased image in a more subtle example for sure.
Now let's take a look at the following somewhat flawed comparison (http://forums.dpreview.com/forums/read.asp?forum=1019&message=30112382) between an SD14 (no AA filter) and a 50D (relatively weak AA filter).
Here is the SD14 100% crop:
(http://thebrownings.name/photo/2010-01-aliasing/106242475.k6guNjIC.SDIM3020-crop.png)
And here is the 50D 100% crop (after downsampling to the same pixel count):
(http://thebrownings.name/photo/2010-01-aliasing/IMG_3026-rawtherapee-4mp-crop.png)
The full SD14 image (http://thebrownings.name/photo/2010-01-aliasing/106242475.k6guNjIC.SDIM3020.jpg)
The full 50D image. (http://thebrownings.name/photo/2010-01-aliasing/IMG_3026-rawtherapee-4mp.png)
In areas where the rivets are close, the 50D shows a slab of fuzzy matter. You will NEVER recover this from sharpening. NEVER, NEVER, NEVER, NEVER, EVER. Period....never.
on the UPRES SD14 image, you STILL get this detail! Amazingly so. (this is in some way a 5-6mp camera) Now the SD14 CLAIMS 14mp res on this camera, but this is with BIG help from the fact that it doesnt have a AA filter! Because if it did, you would see a much fuzzy picture, and they would NOT be able to even make the claim!Period.
-
In areas where the rivets are close, the 50D shows a slab of fuzzy matter. You will NEVER recover this from sharpening. NEVER, NEVER, NEVER, NEVER, EVER. Period....never.
on the UPRES SD14 image, you STILL get this detail! Amazingly so. (this is in some way a 5-6mp camera) Now the SD14 CLAIMS 14mp res on this camera, but this is with BIG help from the fact that it doesnt have a AA filter! Because if it did, you would see a much fuzzy picture, and they would NOT be able to even make the claim!Period.
I looked over the RAW files, and the 50D was not THAT bad....better than the jpegs posted here, even the crops....
BUT...You are comparing a 4.6 vs a 15.1 MP SENSOR!!!
btw...that 50D image had magenta-sickle cell amiba all over...otherwise they need to scrap that entire bridge as it is all rusted...even do a control burn on the forest as it is infected with Magenta-nitus.
thats why these comparison posts are Water under the bridge :-P
-
I don't know where Daniel gets "people prefer the blurred look" data,
Actually, what I said is that most people prefer the aliased version - which is the "un-blurred" look (no AA filter).
And I am not sure how you have quantified your experience?
Sometimes I get this kind of feedback from clients (wedding, portrait, print sales, etc.), but usually it's from conversation with friends and online folks.
How many people have you done this side by side study comparison
I've posted that specific portrait four or five times, but I think there have only a dozen responses or so.
...and why?
To discuss, share, and learn more about aliasing.
Why not see what you like?
I don't understand the question.
-
Hi,
I include two screendumps. Both exposures normalize using "auto" in LR, both white balanced against same spot on bridge.
Two comparisons
1) Canon downscaled with bicubic sharper
2) Sigma upscaled with bicubic softer
The Sigma is here at great disadvantage in 2) because the Canon has much higher resolution. I see less aliasing in the Sigma image than I expected. Downscaling an image is not without perils as it will also introduce some aliasing.
Best regards
Erik
[attachment=19249:Canon_downscaled.jpg]
[attachment=19251:SigmaUpscaled_2.jpg]
In areas where the rivets are close, the 50D shows a slab of fuzzy matter. You will NEVER recover this from sharpening. NEVER, NEVER, NEVER, NEVER, EVER. Period....never.
on the UPRES SD14 image, you STILL get this detail! Amazingly so. (this is in some way a 5-6mp camera) Now the SD14 CLAIMS 14mp res on this camera, but this is with BIG help from the fact that it doesnt have a AA filter! Because if it did, you would see a much fuzzy picture, and they would NOT be able to even make the claim!Period.
-
[quote name='ErikKaffehr' post='338207' date='Jan 7 2010, 02:18 AM']Hi,
I include two screendumps. Both exposures normalize using "auto" in LR, both white balanced against same spot on bridge.
Two comparisons
1) Canon downscaled with bicubic sharper
2) Sigma upscaled with bicubic softer
Best regards
Erik
Nice post to compare Erik!
....wow that SD14 image claim of 14 is not at all an exageration...
I didn't upres the SD14, so seeing it like this (SD14 looks slightly larger?) side to side is very nice.
I see the wires going up showing the jaggedy on BOTH images. I also see some better contrast and slight better detial(?where the fence is on top) on the 50D.
This post shows me how NICE that SD14 really is ! I have much more respect for the Foveon after seeing this closer. Too bad they are maxxed out.
Did I mention 4.6mp no AA vs 15.1mp Canon?!
:-)
Thanks for clarity Daniel...just wanted a bit more on your angle of things....and, so OK, the NO AA is preferred....makes more sense to me :-)
-
Hi,
I can see some of the differences. To me one image is to soft (but I'm no portrait photographer) the other over sharpened. I don't think that those images are shot with different cameras as I cannot see any movement in the object.
I presume that they are created from the same image using different down sampling methods? What about sharpening?
Best regards
Erik
Sure thing. The differences are sometimes very subtle. It's a good idea to first examine an image where the aliasing is very obvious and pronounced. Here is one such demonstration: the first image has lots of aliasing, the second has very little:
(http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-point.png) (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-point.png)
(http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-lanczos.png) (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt-400-lanczos.png)
I suggest opening them in browser tabs for A/B comparison. Here are a few examples of what is different:
- Speckling in the hair (not there in real life).
- Individual hairs of stubble (they should be too small to see).
- Ear is jagged.
- Catchlight zigzags
Those are examples of the kinds of artifacts that make aliased images look fake and unnatural to me.
-
Hi,
I can see some of the differences. To me one image is to soft (but I'm no portrait photographer) the other over sharpened. I don't think that those images are shot with different cameras as I cannot see any movement in the object.
I presume that they are created from the same image using different down sampling methods? What about sharpening?
Best regards
Erik
true observation
-
I am surprised John Sheehy or Graemes Natress have not chimed in....?
All they did was blow some smoke and disapear, with ZERO images to even help their posts. Maybe they are at the drawng board to post some true test images?
And I don't mean it in a negative way...but I hate posts that are so one sided and convinced and there is Little or misguided or even ZERO backup to the claim. So pardon my reactions.
They can easily shut me up, and the others with posting what they find....maybe they just posted textbook/something they read or studied...not experience.
That example of the doctor med percription analogy was very good, I can relate :-)
I hope I am wrong, but the last post with pictures makes lots of sense... 4.6 vs 15.1 WOW...it is a wow for me.
Maybe I need to find someone with a A900 or D3x file so it can go up against the MFdB? I would love to see a SIDE by SIDE RAW file of these 2. I know there was a HOT debate in the MF forum doing a overall comparison...but I dont think on such a specifc area.
-
Hi,
Graeme Natress is doing "raw conversion stuff" for RED, so he does know a couple of things about the issue. Now, RED is about motion pictures and compression utilizing motion prediction. That kind of algorithm really suffers from aliasing artifacts, so I guess Graeme is extra sensitive for it.
Best regards
Erik
I am surprised John Sheehy or Graemes Natress have not chimed in....?
All they did was blow some smoke and disapear, with ZERO images to even help their posts. Maybe they are at the drawng board to post some true test images?
And I don't mean it in a negative way...but I hate posts that are so one sided and convinced and there is Little or misguided or even ZERO backup to the claim. So pardon my reactions.
They can easily shut me up, and the others with posting what they find....maybe they just posted textbook/something they read or studied...not experience.
That example of the doctor med percription analogy was very good, I can relate :-)
I hope I am wrong, but the last post with pictures makes lots of sense... 4.6 vs 15.1 WOW...it is a wow for me.
Maybe I need to find someone with a A900 or D3x file so it can go up against the MFdB? I would love to see a SIDE by SIDE RAW file of these 2. I know there was a HOT debate in the MF forum doing a overall comparison...but I dont think on such a specifc area.
-
I can see some of the differences. To me one image is to soft (but I'm no portrait photographer) the other over sharpened.
The aliasing artifacts in this image are different from over-sharpening artifacts (with the exception that an image that is already aliased will get even worse with sharpening). Try pulling both images into your sharpening workflow and give them the same parameters. I think you'll find that they sharpen much differently: the aliasing artifacts in the first image become much more pronounced with sharpening.
I presume that they are created from the same image using different down sampling methods?
Correct. Both images were generated from the same file:
20 MB original PNG conversion (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481-rt.png)
The PNG was generated from this raw file:
24 MB Original raw file (http://thebrownings.name/photo/2009-10-aliasing/2009-01-30-3481.CR2)
To generate the aliased file, I downsampled the original PNG with a point-sampling algorithm in ImageMagick (a command-line image processing program):
convert -filter Point -resize 400x 2009-01-30-3481-rt.png 2009-01-30-3481-rt-400-point.png
To generate the anti-aliased file, I downsampled the same PNG file with the lanczos algorithm in ImageMagick:
convert -resize 400x 2009-01-30-3481.png 2009-01-30-3481-rt-400-lanczos.png
What about sharpening?
None in either image (not during conversion, post, nor after downsample).
The purpose of the demonstration is to show the difference between lots of aliasing (luma only) and very little aliasing. If someone prefers the aliased image and sees no artifacts in this extreme example, then there is no point in trying to demonstrate the artifacts to them using images with less extreme aliasing. Since Phil did see the artifacts in the extreme example, I moved on to a less extreme example with the SD14. In that image, he did not consider it to have artifacts, so somewhere between the two lies the point at which the aliasing becomes bad enough to cause an issue for him. If instead he had thought that the artifacts in the SD14 image were horrible (as I do), then I would have tried to find a comparison where the aliasing is even less extreme.
I am surprised John Sheehy or Graemes Natress have not chimed in....?
Well, it is a two-year-old thread.
All they did was blow some smoke and disapear, with ZERO images to even help their posts. Maybe they are at the drawng board to post some true test images?
John Sheehy has posted a bunch of images here and on other forums, but I don't think he participates here anymore.
As for Graeme, here are a few examples of his work:
http://www.red.com/shot_on_red/ (http://www.red.com/shot_on_red/)
http://red.cachefly.net/RedReel500.mov (http://red.cachefly.net/RedReel500.mov)
http://red.cachefly.net/redreel/RedReel_h264_720.mov (http://red.cachefly.net/redreel/RedReel_h264_720.mov)
-
Interesting stuff....on the last crps I did see the artifacts...jaggedy effects on the power lines...in BOTH Sd14 and 50D FILES.
And the prev were Jpegs....things got clear with the raw, and the samples gen by Erik...both showing "artifacts" (or were those just jpeg issues? artifacts none the less
And how come we are ignoring the obvious difference in the 2 cameras?
-
I will check those links out for his works. I don't know about moving pictures, as there maybe more to it with image tear and aliasing being more evident...but I will know better after the images I take. and yes my eyes have been recently checked :-)
i'll be back later
-
Did I mention 4.6mp no AA vs 15.1mp Canon?!
:-)
Well, that's the difference between real pixels and sub-pixels in a grid. But both companies make bogus claims.
(excuse the lengthy math here but bear with me)
The Sigma is only 4.6MP. But that's sub APS-C size, or 20.7X13.8mm vs 36X24mm for full-frame. Scaling the sensor up would get you: 2652X1.74 X 1769X1.74 or 4612X3076(14MP) This results in a: 4612/1.4173 (~3250 dpi) equivalent scan of film. Obviously this is a bit overkill, since most film is scanned at 2400dpi and scaling it up like this results in a 20MP scan, essentially. Converting the sensor back to a more reasonable 2400dpi would result in: ~3400X2270.(7.7MP) The Foveon sensor needs to be 26.6X17.7mm.(about the size of the Canon higher-end sensors) This seems reasonable to accomplish from an engineering perspective, but they are relying on marketing nonsense to claim it's "14MP".
Then again, if Sigma did scale it as-is to full frame, it would take a ~32MP Bayer sensor to match it. Since the next generation of DSLRs are going to likely try to break the 30MP barrier, this seems like a no-brainer for Sigma. Just take what you have and scale it to full-frame. Nikon OTOH, has a real issue with trying to cram 30MP in their current top-end sensor. They are hitting some very difficult engineering problems while Sigma just has to make more of the exact same thing to fill the frame.
The other problem is the industry as a whole has confused photoreceptor locations with pixels, so you get idiocy like claiming that the above 12MP camera actually has 12MP. If you scaled to to actual full-color locations, it's really ~4256X0.666 x 2832X0.666, or ~2837X1888 (5.36 million actual full-color locations) This might not sound like a lot, but it's quite respectable, really, if we're talking real MP. But consumers would have a brain fart and not understand because they think bigger is always better. This also shows as well in the pictures. The 12MP Nikon has a slight resolution edge over the SD14. If we were comparing a 10MP Bayer sensor, we'd be roughly equivalent. Except that the AA filter robs the Nikon of a few percentage points to where it's essentially a wash. Neither is 35mm equivalent, but both do a decent job at what they do.
But as a once serious photographer, I don't see either as being enough resolution to do the job. It's probably why I'll just get a cheap A850 and call it a day, since even with the AA filter on it, it more than covers 35mm scanned film.
-
Well, that's the difference between real pixels and sub-pixels in a grid. But both companies make bogus claims.
(excuse the lengthy math here but bear with me)
The Sigma is only 4.6MP. But that's sub APS-C size, or 20.7X13.8mm vs 36X24mm for full-frame. Scaling the sensor up would get you: 2652X1.74 X 1769X1.74 or 4612X3076(14MP) This results in a: 4612/1.4173 (~3250 dpi) equivalent scan of film. Obviously this is a bit overkill, since most film is scanned at 2400dpi and scaling it up like this results in a 20MP scan, essentially. Converting the sensor back to a more reasonable 2400dpi would result in: ~3400X2270.(7.7MP) The Foveon sensor needs to be 26.6X17.7mm.(about the size of the Canon higher-end sensors) This seems reasonable to accomplish from an engineering perspective, but they are relying on marketing nonsense to claim it's "14MP".
The other problem is the industry as a whole has confused photoreceptor locations with pixels, so you get idiocy like claiming that the above 12MP camera actually has 12MP. If you scaled to to actual full-color locations, it's really ~4256X0.666 x 2832X0.666, or ~2837X1888 (5.36 million actual full-color locations) This might not sound like a lot, but it's quite respectable, really, if we're talking real MP. But consumers would have a brain fart and not understand because they think bigger is always better. This also shows as well in the pictures. The 12MP Nikon has a slight resolution edge over the SD14. If we were comparing a 10MP Bayer sensor, we'd be roughly equivalent. Except that the AA filter robs the Nikon of a few percentage points to where it's essentially a wash. Neither is 35mm equivalent, but both do a decent job at what they do.
Makes sense. I wish there was a way to continue the Foveon sensor dev to match up in REAL resolution to the currect 16mp and up cameras...then you would have a serious situation.
I agree the claims are a bit extra....but not too far off . Although in the posts earlier we are comparing the 50D, which is known to be NO BETTER resolution than the 40D (dpreview tests), at ~10MP.
So this makes more sense, as you state the Nik having better res at 12mp.
I really don't care too much about these things....BUT, I DO care that the "REAL" on hands experience so far for me has been a NON issue with artifacts. The crops that were last made by Erik show how....
For the SD14, maybe it is a slight detail that is "smudged" out
For the 50D, It is the entire image that is fuzzy, and making detail ares undistiguishable.
-
Hi,
Just a few comments.
One of the issues is that downsampling creates artifacts. It seems that the routines we have in Photoshop may be less than optimal:
http://www.xs4all.nl/~bvdwolf/main/foto/do...down_sample.htm (http://www.xs4all.nl/~bvdwolf/main/foto/down_sample/down_sample.htm)
On the other hand, Adobe did try Lanzos in Lightroom but decided to go back to a combination of Bicubic and it's versions of smoother and sharper. Testing is often done using patterns that enhance the artifacts. The artifacting may be much less with real life subjects. As I said I'd expect more artifacts in the Foevon image. That said there are areas where I can see an advantage of the Canon in the down sampled comparison.
When the Foevon image is upsampled to the same resolution as the Canon, the Canon image is much better. Make sure that you see the images at "actual pixels".
Best regards
Erik
Makes sense. I wish there was a way to continue the Foveon sensor dev to match up in REAL resolution to the currect 16mp and up cameras...then you would have a serious situation.
I agree the claims are a bit extra....but not too far off . Although in the posts earlier we are comparing the 50D, which is known to be NO BETTER resolution than the 40D (dpreview tests), at ~10MP.
So this makes more sense, as you state the Nik having better res at 12mp.
I really don't care too much about these things....BUT, I DO care that the "REAL" on hands experience so far for me has been a NON issue with artifacts. The crops that were last made by Erik show how....
For the SD14, maybe it is a slight detail that is "smudged" out
For the 50D, It is the entire image that is fuzzy, and making detail ares undistiguishable.
-
I did look at them at actual res. If you look right above the first straight post and other small areas, you see the 50D showing blurr and you cannot see the different rivets, they are a patch area of ?
In the Fov, Yes there are a few Oddly disappear...but, the ones that are there...more than half...are clear and identifiable.
Yes, downsampling, is downsampling, and that is something else..I produce ads, and web is not relevent until we stop printing magazines :-)
which brings me to another reason wanting more snap...clarity...press.
-
It seems that the routines we have in Photoshop may be less than optimal:
Photoshop uses BiCubic among others, which only makes sense if the downsizing ratio is not more than a few pixels to 1 in each dimension. That is why you would see people coming up with heuristics to pre-blur an image before doing downsampling using BiCubic. I don't know if Photoshop does the preblur before BiCubic, but even if it does that is not the best approach. The problem is that stuff such as BiCubic was designed for interpolation (upsampling) and they are more concerned about the keeping the sample values at original pixel locations intact and not designed for information loss, which other techniques such as approximation methods handle gracefully.
It is not fully appreciated many times that the pre-filtering in such methods depends upon the reconstruction method. For e.g., in the case of sinc (ideal) reconstruction the prefilter is again a sinc (ideal low pass filter before sampling). However, if the reconstruction method is not sinc, (say linear, or cubic spline) then the optimal pre-filter is a different one. For e.g., IIRC, for the nearest neighbor reconstruction the optimal pre-filter is the one that averages all signal values in the sampling interval.
-
Well, that's the difference between real pixels and sub-pixels in a grid. But both companies make bogus claims.
The other problem is the industry as a whole has confused photoreceptor locations with pixels, so you get idiocy like claiming that the above 12MP camera actually has 12MP. If you scaled to to actual full-color locations, it's really ~4256X0.666 x 2832X0.666, or ~2837X1888 (5.36 million actual full-color locations) This might not sound like a lot, but it's quite respectable, really, if we're talking real MP. But consumers would have a brain fart and not understand because they think bigger is always better. This also shows as well in the pictures. The 12MP Nikon has a slight resolution edge over the SD14. If we were comparing a 10MP Bayer sensor, we'd be roughly equivalent. Except that the AA filter robs the Nikon of a few percentage points to where it's essentially a wash. Neither is 35mm equivalent, but both do a decent job at what they do.
I think it's a little more complicated than that. If resolution were actually 1.5 times the pixel spacing, one wouldn't be able to resolve parallel lines at Nyquist; but one can, because the information is sampled at the pixel spacing, just in different passbands of the visible spectrum. A good demosaic algorithm uses this fact and reconstructs the lines. Pointlike structures are a different story. So it really depends what sort of signal is presented as to what resolution is achieved; but since many signals are edges and lines rather than fine dots, in those situations something closer to the pixel spacing for resolution is achieved.
-
Hi,
If you look at the images at Sigma native resolution you can see that wires in the sigma (left image have some jaggies):
[attachment=19259:Canon_downscaled.jpg]
If you look at the image at native Canon size you see that the Canon image still has no jaggies (Canon is now on the left side, sorry)
[attachment=19260:SigmaUpscaled_2.jpg]
Best regards
Erik
I did look at them at actual res. If you look right above the first straight post and other small areas, you see the 50D showing blurr and you cannot see the different rivets, they are a patch area of ?
In the Fov, Yes there are a few Oddly disappear...but, the ones that are there...more than half...are clear and identifiable.
Yes, downsampling, is downsampling, and that is something else..I produce ads, and web is not relevent until we stop printing magazines :-)
which brings me to another reason wanting more snap...clarity...press.
-
Hi,
If you look at the images at Sigma native resolution you can see that wires in the sigma (left image have some jaggies):
[attachment=19259:Canon_downscaled.jpg]
If you look at the image at native Canon size you see that the Canon image still has no jaggies (Canon is now on the left side, sorry)
[attachment=19260:SigmaUpscaled_2.jpg]
Best regards
Erik
Yes even in the thicker cable areas, you see the line in better seperation....Eh Allora?!? 4.6 vs 15.1
My post is not Sd14 vs the claim of 14mp...allora
-
Hi,
Yes of course, but we see no jaggies at actual pixels. Three possible explanations
1) Detail is significantly thicker den pixel size, so it covers more then one pixel
2) Sensor outresolves lens, or more correctly, the lens has low MTF at double Nyquist frequency
3) AA-filter is doing the job it is intended to do
Or a combination of the three factors above
Best regards
Erik
Yes even in the thicker cable areas, you see the line in better seperation....Eh Allora?!? 4.6 vs 15.1
My post is not Sd14 vs the claim of 14mp...allora
-
Hi,
Yes of course, but we see no jaggies at actual pixels. Three possible explanations
1) Detail is significantly thicker den pixel size, so it covers more then one pixel
2) Sensor outresolves lens, or more correctly, the lens has low MTF at double Nyquist frequency
3) AA-filter is doing the job it is intended to do
Or a combination of the three factors above
Best regards
Erik
Thank you Erik for helping understand these things in a plain fashion as I am not well versed in the scientific terms.
The possibility of the line on a single pixel is something that comes to my mind.
So if the SD14 had higher resolution, would we be seeing those jaggy lines (also see it on the thicker lines going diagonally up) ?
I am not sure of #2 being a factor, it is beyond my observations and understanding of the relation.
#3, I would like to know if the 50D has a lower RAW setting...ONLY THEN can we see if the AA is doing the job...As I suspect the line would be a blurry darker tone of NOTHING if it were the case. But I cannot say for sure. :-)....
Because we have to calculate the measure of how it is viewed.
-
I don't know about moving pictures, as there maybe more to it with image tear and aliasing being more evident...
Glad you mentioned moving pictures. One may be surprised that the aliasing offered by Bayer sensor is very similar in linear theoretical treatment to aliasing in an NTSC TV signal. At least 25 years ago before Shannon proposed a simple theoretical way of sampling using a lowpass filter, TV engineers were aware of aliasing and figuring out how to remove it. Shannon himself acknowledged this fact in his landmark paper. However, the early TV engineers termed "aliasing" as "confusion" in a signal, and it was of a very interesting variety as a 2D picture was scanned like a 1-D signal. However, in any case, when chroma info aliased into luma it produced those "crawling dots", and when luma aliased into chroma it produced those dreaded "moving color patterns". Among the artifacts offered by linear analysis of Bayer CFA, the luma and chroma aliasing happen almost exactly in the same manner.
-
However, in any case, when chroma info aliased into luma it produced those "crawling dots", and when luma aliased into chroma it produced those dreaded "moving color patterns".
Among the artifacts offered by linear analysis of Bayer CFA, the luma and chroma aliasing happen almost exactly in the same manner.
Knowing the "contamination" is interesting to me...at least from what you say. So was Bayers analysis proven? The lum/Chrom alias happen in the same manner as ? (Bayer makes moving picture sensors)
So Shannon "invented" the AA filter?
-
Knowing the "contamination" is interesting to me...at least from what you say. So was Bayers analysis proven? The lum/Chrom alias happen in the same manner as ? (Bayer makes moving picture sensors)
Theoretical treatment of aliasing of luma and chroma signal (for NTSC defined as YCbCr/YIQ) and for Bayer CFA (defined as a certain linear combination of RGB) are very similar.
So Shannon "invented" the AA filter?
No it wouldn't appear that Shannon invented the AA filter. The ideas were known earlier. In fact the term Nyquist frequency is coming from some idea from a paper by Nyquist in late 1920's or 1930's IIRC, where as Shanon presented his paper in late 1950's. Whitaker had similar ideas as early as 1904. And some Russian scientists (Kotelnikov) were aware of and presented analysis much earlier than Shannon. What Shannon did was to outline a simple theoretical way of sampling without aliasing by incorporating a lowpass filter. Therefore, of course, Shannon's sampling theorem applies to bandlimited signals. However, even Shannon's ideas have now been surpassed and they are only considered a special case in the more general non-bandlimited sampling (from L2 space) to other spaces.
The thing is people have invented same things over and over again. Think wavelets! Those were good days when Engineering had connection with Mathematics and Physics. These days for many areas in Electrical Engineering (graduate level) Physics is largely irrelevant, which is sad.
-
so this is film techniques and formulas applied to digital...surely many things can be learned. thanks.
-
Theoretical treatment of aliasing of luma and chroma signal (for NTSC defined as YCbCr/YIQ) and for Bayer CFA (defined as a certain linear combination of RGB) are very similar.
You keep going on and on about this and how aliasing can be removed without altering any image detail. How about posting a link to a scientific paper or something instead of repeating the same assertions over and over without offering a single shred of evidence to back them up?
-
You keep going on and on about this and how aliasing can be removed without altering any image detail. How about posting a link to a scientific paper or something instead of repeating the same assertions over and over without offering a single shred of evidence to back them up?
You see what happen Jonathan...:-)
So far we have 7 pages, and really NO apple to apple comparison of a AA vs NON-AA of "artifacts" coza che?
Apple to Apple would be acceptable for....
1.both sensors to be in the ball park...so 1 to 2 difference is not a big issue, as we already are splitting hairs.
2. same lens...exact actual physical lens. for obvious reasons...(I know, nothing is obvious so far)
3. Same, unmoved or altered still subject under exact same TESTED light. (we don't want any inconsistant pops)
4. ?? feel free to add something worth the digital space on MR MR's servers.
It will take me a bit longer to do the test. I have plenty of NON AA images I can post, but my other Bodies are not going to be back for a bit longer than I though (next week sometime).
So in the meantime, I would love to see any other comparisons...and please state if the images comply to the few restrictions, or if they are NOT, but somewhat close and worth a look.
-
Hi,
To begin with:
Thinks about two horizontal rows of pixels with a line nearly parallel to pixels rows. If the pixels are small with spacing between them the line could either affect the upper row or the lower one. So the line would at one point flip from one row to next one. This is called the staircase effect some times. Now if the pixels are large, the line line would affect both rows simultaneously at a stage, so it would glide from one row to another. A thin line would still flip. I'm not good at illustration but you could try it with a ruler and checked paper.
As long a lens can project a line that is thinner than the spacing between the sensors this problem will arise.
Now, what is resolution? Resolution is the ability to tell small thing at small distances apart. To be able to do that you need some contrast (like a dark line between two bright lines). MTF just measures the contrast between the dark and light part of a line pair. The thinner the line pairs are the lower the contrast will be. Ultimately MTF and resolution will be limited by diffraction.
Now, let's consider the Sigma with a pixel spacing of about 0.08 mm. It would have about 63 lp/mm. At this frequency simpler lenses will transfer really little contrast. Most MTF charts cover 10, 20 and 40 lp/mm, so we are essentially left with very little data. A perfect lens may have something like 65% contrast at f/8. This says essentially that with a lower quality lens a line pair would be mostly gray, with a truly excellent lens they would be dark gray and very light gray.
So a good lens would show jaggies while a less good lens would smear out the lines.
A Swedish monthly that tests both cameras and lenses has discovered that they need to find optimal apertures for resolution tests on new cameras. Earlier they could just use top notch optics stopped down to f/8 but this is clearly not the case any longer. For 4/3 cameras and also the Canon 50D they need about the best lens in the inventory, and stop it down to f/5.6 resolution worsens by f/8. Now, the Swedish monthly is using Imatest and what they actually measure is the resolution where half the contrast is lost, and I think this also takes some sharpening into account. But there is no doubt that sensor pitches are close to be limited by lens performance.
You will find absurdly high resolution figures for lenses like 200 lp/mm, but those figures are only possible at large apertures an at smallish contrast.
This figure that I have measured myself illustrates the issue.
http://83.177.178.7/ekr/images/stories/difractionlimit.gif (http://83.177.178.7/ekr/images/stories/difractionlimit.gif)
2000 Lines/PH corresponds to about 63 lp/h. You can see that resolution is diminishing when stopping down beyond f/8. Better lenses would shift the peak to the left. Higher resolution would push the peak upwards. Note that the Alpha 100 (at 10 MP) doesn't resolve much higher than the KonicMinolta 7D (at 6 MP) past f/11 or so. This limitation is by laws of physics, it applies to Leica and Zeiss, too. For optimum sharpness you cannot stop down!
Best regards
Erik
Thank you Erik for helping understand these things in a plain fashion as I am not well versed in the scientific terms.
The possibility of the line on a single pixel is something that comes to my mind.
So if the SD14 had higher resolution, would we be seeing those jaggy lines (also see it on the thicker lines going diagonally up) ?
I am not sure of #2 being a factor, it is beyond my observations and understanding of the relation.
#3, I would like to know if the 50D has a lower RAW setting...ONLY THEN can we see if the AA is doing the job...As I suspect the line would be a blurry darker tone of NOTHING if it were the case. But I cannot say for sure. :-)....
Because we have to calculate the measure of how it is viewed.
-
so this is film techniques and formulas applied to digital...surely many things can be learned. thanks.
True. Analog to digital techniques always are a kludge, and that's the main problem with the Bayer pattern - it requires a bunch of massaging of the data to deal with the problems.
Yes of course, but we see no jaggies at actual pixels. Three possible explanations
The reason is because the Foveon sensor bleeds between each pixel location like film or dye-sub printers do(almost zero space between the locations), so unless it's a hard line, you'll never likely notice it. OTOH, if you look at a dark scene with sunlight in it(say under a bridge looking out at the lighter distance), the Foveon sensor has horrendous *everything* that you can think of along that line where the two meet. This is exacerbated by the low DR as well. Its very sensitive to white levels, but if that's even close or there aren't hard to deal with issues, it's awesome. It, IMO, is the best sensor made to date for outdoor scenery. Kind of like having Velvia 50 always in your camera and never using anything else. Nothing better when it's in its element but pretty much give up using it for interiors or low light barring a tripod and a lot of pushing the exposure.(and praying as any SD series user will attest to - heh )
IMO, the Fuji is the better compromise. Crazy high DR and film like shoulder, plus very little noise and artifacts. I originally was THE biggest Foveon fan out there but the sobering truth is that you can do so much more with the Fuji. If only it wasn't like a... well, it's kind of like giving a person an eye dropper of hundred year old Brandy. It's almost cruel in a way... Double the MP would be enough, and full frame would just about do that for them.
For the SD14, maybe it is a slight detail that is "smudged" out
For the 50D, It is the entire image that is fuzzy, and making detail ares indistinguishable.
As for this comment, it's actually the reverse. The reason the SD14 looks smudged out is because it's hit a hard wall. The locations are spaced right next to each other like film and so when it hits that limit, it resolves nothing more. It's not smudged but is suffering from myopia/is truly fuzzy. The more you push the Foveon as a result, the less in-focus tiny details become. But it still blurs consistently as every color channel is identical in resolution.
The 50D is suffering from different channels having different resolution. This causes horrendous aliasing and artifacts in tiny details. The software tries it butt off to fix this, but it's not truly fixable unless the pixels are so tiny that single points of garbage are too small to see individually(which the 24MP cameras come close to, btw) The end result is that the camera smudges the detail to smooth it out. So areas look great and areas also look bad in the same image.
There is no free lunch. Both technologies have things that you love and hate.
-
Hi Erik,
this last post on my first read has a lot of meat in it. I think I will need to read it 2 or 3 times to graps it well enough. So far the first part I got, not a problem...I understand it
I will need to read the rest with more care. I am following, but need a better understanding.
I know deffraction as I have many times been puzzled, then I understood. But what you say regarding the spacing will vary from each camera model , i would think.
So if we know the spacing of the Kodak SLRc, we can match it with a lens so we get the zigzag instead of the smear...As I think the zigzag maybe easier to handle?
The interesting thing is the 50D has NO space between the microlens. although these are atop the photons, or what ever they are called...maybe that is still the issue.
Lots to think about...BUT
In my experience I like the aliasing file, as I think it is the AA blurring that takes away from the "polish" look, even with the zigzag on occasion...Besides, it is NOT the aliasing that makes it look good. Because Daniel, you explanation of "people liking it" sounds like it is the aliasing they like....NO, it is the lack of the AA they like....I think.
OK, I will read a few more times later and post. :-)
-
True. Analog to digital techniques always are a kludge, and that's the main problem with the Bayer pattern - it requires a bunch of massaging of the data to deal with the problems.
The reason is because the Foveon sensor bleeds between each pixel location like film or dye-sub printers do(almost zero space between the locations), so unless it's a hard line, you'll never likely notice it. OTOH, if you look at a dark scene with sunlight in it(say under a bridge looking out at the lighter distance), the Foveon sensor has horrendous *everything* that you can think of along that line where the two meet. This is exacerbated by the low DR as well. Its very sensitive to white levels, but if that's even close or there aren't hard to deal with issues, it's awesome. It, IMO, is the best sensor made to date for outdoor scenery. Kind of like having Velvia 50 always in your camera and never using anything else. Nothing better when it's in its element but pretty much give up using it for interiors or low light barring a tripod and a lot of pushing the exposure.(and praying as any SD series user will attest to - heh )
IMO, the Fuji is the better compromise. Crazy high DR and film like shoulder, plus very little noise and artifacts. I originally was THE biggest Foveon fan out there but the sobering truth is that you can do so much more with the Fuji. If only it wasn't like a... well, it's kind of like giving a person an eye dropper of hundred year old Brandy. It's almost cruel in a way... Double the MP would be enough, and full frame would just about do that for them.
As for this comment, it's actually the reverse. The reason the SD14 looks smudged out is because it's hit a hard wall. The locations are spaced right next to each other like film and so when it hits that limit, it resolves nothing more. It's not smudged but is suffering from myopia/is truly fuzzy. The more you push the Foveon as a result, the less in-focus tiny details become. But it still blurs consistently as every color channel is identical in resolution.
The 50D is suffering from different channels having different resolution. This causes horrendous aliasing and artifacts in tiny details. The software tries it butt off to fix this, but it's not truly fixable unless the pixels are so tiny that single points of garbage are too small to see individually(which the 24MP cameras come close to, btw) The end result is that the camera smudges the detail to smooth it out. So areas look great and areas also look bad in the same image.
There is no free lunch. Both technologies have things that you love and hate.
One of your better posts that help me understand :-)
thank you
-
so this is film techniques and formulas applied to digital...surely many things can be learned. thanks.
I don't want to give an impression that there is some analog formula that gets applied to digital. Fourier transform of sampled system (discrete system) is continuous and periodic, where as the way 2D->1D Fourier transform is taken in NTSC TV analog signal it makes it (approximately) discrete in the frequency domain. Now isn't that interesting that a discrete system has a continuous waveform in frequency domain and an analog signal treated in a certain way has a discrete representation. And, there is a close relationship between such continuous/discrete representations in the frequency domain. Don't let the Discrete Fourier Transform (DFT) of a discrete signal confuse you as it assumes certain periodicities in the spatial domain that makes it discrete in the frequency domain, and again there is a close relationship of the DFT to above mentioned continuous/discrete frequency representations.
Anyway, I'm getting boring here so lets change gears, and suffice it to say once you bring stuff into frequency domain then the usual notion of aliasing starts making sense and one sees that luma/chrom leaking (aliasing) into each other for both analog and discrete systems results in similar observations.
Bayer CFA may be considered as a combination of luma and 2 chroma components derived using a certain linear combination of RGGB CFA values. NTSC TV signal also has a luma and 2 chroma components (YIQ/YUV), but of course derived using different weights in the linear combination of RGB. And, these nasty beasts leak (alias) into each other resulting in luma/chroma aliasing.
For e.g., haven't we seen those "color fringing" in digital photos for high detail scenes (luma aliasing into chroma). Now compare to those "color patterns" that result in NTSC TV for high detail scenes. When chroma aliases into luma the digital photos start having those dots, where as NTSC TV signal has those "crawling dots".
-
Anyway, I'm getting boring here so lets change gears, and suffice it to say once you bring stuff into frequency domain then the usual notion of aliasing starts making sense and one sees that luma/chrom leaking (aliasing) into each other for both analog and discrete systems results in similar observations.
Bayer CFA may be considered as a combination of luma and 2 chroma components derived using a certain linear combination of RGGB CFA values. NTSC TV signal also has a luma and 2 chroma components (YIQ/YUV), but of course derived using different weights in the linear combination of RGB. And, these nasty beasts leak (alias) into each other resulting in luma/chroma aliasing.
For e.g., haven't we seen those "color fringing" in digital photos for high detail scenes (luma aliasing into chroma). Now compare to those "color patterns" that result in NTSC TV for high detail scenes. When chroma aliases into luma the digital photos start having those dots, where as NTSC TV signal has those "crawling dots".
Ya, that first part...I am sure other with some insight appreciate it, but....
So the Luma Chroma interesting...
But lets get off the history lesson, and try to see if there is anything we can conclude....
So far NOT.
We DO know that this artifact occurs....Now to see how significant it is. And can we not consider the effects of the AA a imposed artifact?
and lastly....
how these 2 plays into photography in the hands of someone who knows what they are doing?
I honestly see little importance to this in 50% of shooting in the "PROFESSIONAL" level (excluding event shooters, and other mediums that have no need for extreme detailed images).
I believe this is interest more towards...
Firstly a very fussy group. A group that always questions themselves and how they can improve their work, their gear, etc...Looking to always do better.
These are good things I believe, so Then there are studio shooters of cars, or jewelry, of product, of faces, of landscapes, of scientific areas, that this DOES play a somewhat IMPORTANT part of imagery....Of course this is all AFTER you have your skill as a shooter as second nature).
-
One thing to point out is that aliasing maybe gets a worse rap than it deserves.
Perhaps some people like aliasing. Images produced by cameras in the first place are not accurate to the way normal vision perceives images.
I like a certain amount of aliasing for some of my work.
-
One thing to point out is that aliasing maybe gets a worse rap than it deserves.
Perhaps some people like aliasing. Images produced by cameras in the first place are not accurate to the way normal vision perceives images.
I like a certain amount of aliasing for some of my work.
Exactly, so lets see a couple tests, and see how much it bothers the visual. So far the very thin vertical line is a NON issue for me (The one Erik points out). but if at 100% the other thicker line...if that is visible to the eye as jigjag, then it maybe something to treat in PS. If that is ALL the little artifacts that aliasing does, Perhaps it isn't a issue. And I want to see a example where aliasing is preferred?
I think the entire picture is what looks better, and perhaps not the alaising in particular areas. So far more people like to see a NON AA filter according to Damiel, and I think it makes sence, since everything is more clear, with a couple slight issues in a few detail spots. Maybe hair in a portrait might be a problem....we can only see results from images taken. Not becasue aliasing occures, therefore we should use AA which in my observations degrades the ENTIRE image.
-
Hi,
I'd like to point out some issues I think you miss.
1) Resolution on 24 MP cameras is essentially like 10 MP on APS-C. The 5D and the 7D get much closer to oversampling than any any other DSLR, except for 4/3. I presume that you talk about full frame regarding 24 MP because there are no 24 MP APS/C cameras. MFDBs are a different thing.
2) The bayer pattern is a pretty close approximation of the fovea. Our vision is based on color sensitive cones, which at least have some similarity i to the "Bayer" patten. I'm not advocating that mother nature is always correct, but a "Bayer pattern" like solution was choosen by it.
3) Half of the "Bayer pattern" pixels are green, so half of the luminance information is non interpolated. Once the images are converted to JPEG much of the color information is disposed of anyway, so I'd guess that once we have JPEG the amount of chroma information may be similar between "Foveon" and "Beyer".
There is one obvious advantage of Foveon over Bayer, and that is that color Moiré patterns would not arise with Foveon. For that reason Foveon can be used without an AA filter, without artifacting being obvious. Monochrome aliasing in sensors depends only pixel pitch, fill factor and AA-filtering. As Foveon sensors have low Pixel count (that is large pitch) it is obvious that they would alias more than smaller pitch sensors having similar fill factors.
My guess is that the Foveon concept may have a problem with noise and color reproduction. Bayer solutions use filters arrays with spectral characteristics that can be optimized . The Foveon concept is based on diffusion of different wavelengths to different depth. Therefore I presume that it's capability to differentiate colors is somewhat limited. It is quite possible that Foveon has not move to increased resolution because of issues with noise at high ISOs
There may be be two reasons that the "Foveon" concept did not catch on:
1) It may be that the disadvantages outweight the advantages
2) Intelectual propert issues
Best regards
Erik Kaffehr
True. Analog to digital techniques always are a kludge, and that's the main problem with the Bayer pattern - it requires a bunch of massaging of the data to deal with the problems.
The reason is because the Foveon sensor bleeds between each pixel location like film or dye-sub printers do(almost zero space between the locations), so unless it's a hard line, you'll never likely notice it. OTOH, if you look at a dark scene with sunlight in it(say under a bridge looking out at the lighter distance), the Foveon sensor has horrendous *everything* that you can think of along that line where the two meet. This is exacerbated by the low DR as well. Its very sensitive to white levels, but if that's even close or there aren't hard to deal with issues, it's awesome. It, IMO, is the best sensor made to date for outdoor scenery. Kind of like having Velvia 50 always in your camera and never using anything else. Nothing better when it's in its element but pretty much give up using it for interiors or low light barring a tripod and a lot of pushing the exposure.(and praying as any SD series user will attest to - heh )
IMO, the Fuji is the better compromise. Crazy high DR and film like shoulder, plus very little noise and artifacts. I originally was THE biggest Foveon fan out there but the sobering truth is that you can do so much more with the Fuji. If only it wasn't like a... well, it's kind of like giving a person an eye dropper of hundred year old Brandy. It's almost cruel in a way... Double the MP would be enough, and full frame would just about do that for them.
As for this comment, it's actually the reverse. The reason the SD14 looks smudged out is because it's hit a hard wall. The locations are spaced right next to each other like film and so when it hits that limit, it resolves nothing more. It's not smudged but is suffering from myopia/is truly fuzzy. The more you push the Foveon as a result, the less in-focus tiny details become. But it still blurs consistently as every color channel is identical in resolution.
The 50D is suffering from different channels having different resolution. This causes horrendous aliasing and artifacts in tiny details. The software tries it butt off to fix this, but it's not truly fixable unless the pixels are so tiny that single points of garbage are too small to see individually(which the 24MP cameras come close to, btw) The end result is that the camera smudges the detail to smooth it out. So areas look great and areas also look bad in the same image.
There is no free lunch. Both technologies have things that you love and hate.
-
There is one obvious advantage of Foveon over Bayer, and that is that color Moiré patterns would not arise with Foveon. For that reason Foveon can be used without an AA filter, without artifacting being obvious.
I disagree. The aliasing is certainly still "obvious" -- the only question is whether you consider that obvious part of the image as a hideous artifact or beautiful detail. The Foveon image I posted above has only luma aliasing, and to me it is horribly disfigured with artifacts. Yet Phil looks at the same image and he feels that it is highly attractive with sharpness.
If what you meant is that luma aliasing is less ugly than chroma aliasing, I would agree. I think a lot of people who are in love with luma aliasing actually dislike chroma aliasing. (That's why my first demonstration in this thread only had luma aliasing.)
Furthermore, there may be another factor that you did not consider. In most photos taken with filtered Bayer images, chroma aliasing is not a big issue. It's only certain subjects and conditions where the chroma aliasing becomes as objectionable as unfiltered cameras (which have both strong luma and chroma aliasing). In those conditions, it's possible to use a demosaic algorithm that relies much more heavily on the anti-aliased luma signal to avoid aliasing in the chroma signals. The trade off is reduced resolution (especially chroma resolution). Graeme Nattress did this for RED's raw converter.
In other words, the filtered Bayer chroma aliasing problem is well-resolved for all but the most extreme corner cases, making it much less of a weakness in comparison to alternatives like Foveon. (The most important weakness of Bayer is the loss of so many stops of light to color filters, but so far no alternative has proven superior in this regard, including Foveon. Maybe microprisms some day...)
-
Hi,
I'd suggest that it is pretty obvious that luma aliasing is not obvious to quite a few observers, at least judging from the writing on these forums.
Regarding the loss of light in filters I'd guess it is unavoidable as colored light needs to split up so it can be detected by different sensors. As long as filtering is involved light will be lost.
Best regards
Erik
I disagree. The aliasing is certainly still "obvious" -- the only question is whether you consider that obvious part of the image as a hideous artifact or beautiful detail. The Foveon image I posted above has only luma aliasing, and to me it is horribly disfigured with artifacts. Yet Phil looks at the same image and he feels that it is highly attractive with sharpness.
If what you meant is that luma aliasing is less ugly than chroma aliasing, I would agree. I think a lot of people who are in love with luma aliasing actually dislike chroma aliasing. (That's why my first demonstration in this thread only had luma aliasing.)
Furthermore, there may be another factor that you did not consider. In most photos taken with filtered Bayer images, chroma aliasing is not a big issue. It's only certain subjects and conditions where the chroma aliasing becomes as objectionable as unfiltered cameras (which have both strong luma and chroma aliasing). In those conditions, it's possible to use a demosaic algorithm that relies much more heavily on the anti-aliased luma signal to avoid aliasing in the chroma signals. The trade off is reduced resolution (especially chroma resolution). Graeme Nattress did this for RED's raw converter.
In other words, the filtered Bayer chroma aliasing problem is well-resolved for all but the most extreme corner cases, making it much less of a weakness in comparison to alternatives like Foveon. (The most important weakness of Bayer is the loss of so many stops of light to color filters, but so far no alternative has proven superior in this regard, including Foveon. Maybe microprisms some day...)
-
I disagree. The aliasing is certainly still "obvious" -- the only question is whether you consider that obvious part of the image as a hideous artifact or beautiful detail. The Foveon image I posted above has only luma aliasing, and to me it is horribly disfigured with artifacts. Yet Phil looks at the same image and he feels that it is highly attractive with sharpness.
Furthermore, there may be another factor that you did not consider. In most photos taken with filtered Bayer images, chroma aliasing is not a big issue.
In other words, the filtered Bayer chroma aliasing problem is well-resolved for all but the most extreme corner cases, making it much less of a weakness in comparison to alternatives like Foveon.
You guys bring up the chroma to luma differences of aliasing, that is a nice thing for the labs to work on and programmers to control. surely good for us all to read and better understand.
Daniel....You are taking things out of contex and amplifying the strengths of BAYER...why?
Photography as a art, or visual medium is eventually the observers domain, and the behind the scenes, digital of how it is presented is very much scientific....(this is loosely stated, as there are scientific and visual benifits as I see to not having AA even with so far the single image we have "inspected")....
Ultimately, the NON AA sensor, looks BETTER, and the aliasing is NOT obvious, as in first you quote the word "obvious" to Cover your a$s, and then use the word to down play the significance. I don't appreciate that type of language when science is concerned. You become a tech sales support specialist, and help sell a product, OR drive your own agenda.... but not resolve a question.
I appreciate your input: I would more so appreciate a agressively neutral stance and open to what this all means to Photography and visual appeal.
I will post a few pictures ASAP, of NON AA. ANY subject you(anyone), would like me to shoot?
-
I'd suggest that it is pretty obvious that luma aliasing is not obvious to quite a few observers, at least judging from the writing on these forums.
I did not make myself clear. I agree that many observers consider luma aliasing to be highly desirable, so they do not see it as an image riddled with artifacts.
Daniel....You are taking things out of contex [...]
Sorry, I can't see where I did that. I did not mean to.
[...] and amplifying the strengths of BAYER...why?
I don't think I'm amplifying anything. I simply responded to Erik's assertion that Foveon has an advantage over Bayer when it comes to chroma aliasing. I explained why I think that's only a small advantage.
Ultimately, the NON AA sensor, looks BETTER, and the aliasing is NOT obvious, as in first you quote the word "obvious" to Cover your a$s, and then use the word to down play the significance. I don't appreciate that type of language when science is concerned.
Sorry, but I don't understand your rant here at all. Maybe it will help if I break it down:
- Erik said "Foveon can be used without an AA filter, without artifacting being obvious".
- I felt that the statement might mislead a reader into thinking that it will not have strong aliasing.
- If Sigma used Foveon with an AA filter (as recommended by the Foveon designers themselves), then one could also say it will not have obvious artifacts.
- So what is the difference between "Foveon with filter has no obvious artifacts" and "Foveon without filter has no obvious artifacts"?
- The difference is that many do not consider luma aliasing to be an artifact.
- So again, Unfiltered Foveon does have obvious aliasing. That is a fact.
- Some consider aliasing to be beautiful and desirable details.
- Others consider aliasing to be ugly and horrible artifacts.
- To clear up this situation, I said "The aliasing is certainly still "obvious" -- the only question is whether you consider that obvious part of the image as a hideous artifact or beautiful detail."
You become a tech sales support specialist, and help sell a product, OR drive your own agenda.... but not resolve a question.
My only agenda is to learn and share about photography.
-
Hi,
I'd like to point out some issues I think you miss.
1) Resolution on 24 MP cameras is essentially like 10 MP on APS-C. The 5D and the 7D get much closer to oversampling than any any other DSLR, except for 4/3. I presume that you talk about full frame regarding 24 MP because there are no 24 MP APS/C cameras. MFDBs are a different thing.
Correct. I was talking about what it would take to match the Foveon if it were scaled to FF. As it is, it's roughly 10-12MP, and that was fantastic four years ago or so, but now you see pocket consumer cameras with 10-12MP. That are 5-10X mroe usable since they actually do indoor and night-time shots unlike the SD series.
It's kind of like giving a flawless diamond to your fiance'. But 1/8th of a carat. It's awesome to look at, but she's always wondering about the size. Sometimes bigger and slightly less quality is better.
-
My only agenda is to learn and share about photography.
I'm glad to hear that.
I think we have exhausted the ideas, the science, the theories and just about all other measure, except for a mix of image samples of different subjects without an AA that exhibits artifacts that are disruptive of the image.
So far we have viewed a very extreme or exagerated image of aliasing, Danial's posted image. This helps us better define ...aliasing/artifact (I see that aliasing is an artifact, but not really synonimous).
Second images....lead us to take a different turn into comparing apples and oranges....We put some energy into this, but at the end of the day, we beat up a dead horse...sort of. At least it create more dialogue to understand a few things better.
So...We need NON AA images, and I will take some tomorrow. I just wont have a equal or close to equal samples of AA filter image, as part of my gear is in service.
-
It's kind of like giving a flawless diamond to your fiance'. But 1/8th of a carat. It's awesome to look at, but she's always wondering about the size. Sometimes bigger and slightly less quality is better.
Thats a rather good analogy, as "artifacts" and inclusions are sometimes not visible unless you use a microscope and a color index :-) BUT there is a science that will define the stone as a VS1 or VVS1, and mark it as a second verses a Flawless...and worse yet, if you have one with much flourescence that is invisible to the naked eye, it will mark it down also. Then color...hahaha...much indeed a good analogy. :-)
-
Ultimately, I would very much contemplate having the AA filter removed if it became possible to do that. To me, these clinical argument for the AA filter do not go along with real life and remind me very much of my doctor than can only prescrible high priced chemical medicines and balks if i suggest any natural alternatives. For him, there is only one way, but I know that is not true.
So i need to decide between "power files" of the sMK3 or the cleaner higher ISO and clearer pictures of the D3. I lean towards the sMK3 and the luminous sharpening in PK may tip the balance.
Peter
First of all a doctor (I am an RN) cannot or at least should not recommend "natural medicines" for which there is no scientific evidence regarding their efficacy. Because they are marketed as "nutritional" they are not required to undergo testing, but they should be. I think the idea of removing the AA filter is a good one for you, as you say AA doesn't bother you. Different strokes for different folks.
Be careful not to poison yourself with toxic plants or AA though.
-
I think the idea of removing the AA filter is a good one for you, as you say AA doesn't bother you. Different strokes for different folks.
Be careful not to poison yourself with toxic plants or AA though.
We build custom cameras for many medical imaging purposes such as fluorescence microscopy, etc. Many of our customers prefer not having an AA filter. As I have mentioned before this aliasing thingy is in the notice of engineers for at least 80 years. It is not a new thing and in natural images not a big deal as some camera manufacturers would like to make it who want to tout their superior pixel count. Several new and exciting technologies are going to hit the imaging world and I think it is about time if we find aliasing to be too objectionable to just use a camera which has an AA filter and move on.
-
We build custom cameras for many medical imaging purposes such as fluorescence microscopy, etc. Many of our customers prefer not having an AA filter.
This by itself doesn't say much. If the subject matter has low contrast, and the optical chain functions as a low-pass filter, then adding an AA-filter would be overkill.
... I think it is about time if we find aliasing to be too objectionable to just use a camera which has an AA filter and move on.
And the majority of cameras do use an AA-filter (or an optical chain that functions as such). Some subject matter is not sensitive to aliasing artifacts, and some is. It's a matter of how are you going to cope with it when you get bitten by it. It's all a trade-off.
That being said, the losses of a sufficiently high resolution sampling with an optical low-pass filter are quite tolerable (and in part reversible by proper sharpening), and the amount of nasty surprises is reduced. Not a bad compromise I would say.
Cheers,
Bart
-
This by itself doesn't say much. If the subject matter has low contrast, and the optical chain functions as a low-pass filter, then adding an AA-filter would be overkill.
Hi, I am just saying that since, you, me, Daniel Browning, and countless others agree that aliasing is a subjective matter of appreciation then we are spending a disproportionate amount of time on its ill-effects.
-
That being said, the losses of a sufficiently high resolution sampling with an optical low-pass filter are quite tolerable (and in part reversible by proper sharpening), and the amount of nasty surprises is reduced. Not a bad compromise I would say.
Cheers,
Bart
NO, you cannot regain lost sharpening, you will only increase contrast, and preceived sharpness, no matter the channel, it is applying the same effect.
Some result better than others, but even with some samples posted here, you will see that you CANNOT regain the information that native sharpness resolved.
-
Hi,
It's actually quite possible to regain information if the Point Spread Function (PSF) is known.
Not that I have expertise in the area, but you may check "Digital Image Processing" by Gonzales and Woods or google on "PSF deconvolution Richardson Lucy" and you ought to find examples.
An example is here: http://www.bialith.com/Research/BARclockblur.htm (http://www.bialith.com/Research/BARclockblur.htm)
Best regards
Erik
NO, you cannot regain lost sharpening, you will only increase contrast, and preceived sharpness, no matter the channel, it is applying the same effect.
Some result better than others, but even with some samples posted here, you will see that you CANNOT regain the information that native sharpness resolved.
-
Hi,
It's actually quite possible to regain information if the Point Spread Function (PSF) is known.
Not that I have expertise in the area, but you may check "Digital Image Processing" by Gonzales and Woods or google on "PSF deconvolution Richardson Lucy" and you ought to find examples.
An example is here: http://www.bialith.com/Research/BARclockblur.htm (http://www.bialith.com/Research/BARclockblur.htm)
Best regards
Erik
Erik, it would be great to see the revits reappear from the 50D image crop you posted. The SD14 shows them nice and seprate. Perhaps a visual would be most convincing. I have invested too much time with all this and so far ZERO visual support[size="2"][/size]. Also, there are I am sure examples that might be able to do this as calculations are obvious ....but the tiny revits is a good example to try and REVIVE!
-
Phil,
I don't understand the question. If you mean the rivets I don't really know which rivets you mean, If you could encircle that part on one of the original photographs I could see what you mean.
Best regards
Erik
Erik, it would be great to see the revits reappear from the 50D image crop you posted. The SD14 shows them nice and seprate. Perhaps a visual would be most convincing. I have invested too much time with all this and so far ZERO visual support[size="2"][/size]. Also, there are I am sure examples that might be able to do this as calculations are obvious ....but the tiny revits is a good example to try and REVIVE!
-
Phil,
I don't understand the question. If you mean the rivets I don't really know which rivets you mean, If you could encircle that part on one of the original photographs I could see what you mean.
Best regards
Erik
Yes, there is a recatangle shape area that has rivets in it. the difference is really not much at all as this again is 4.6 vs 15.1mp.
But I am taking a few images now with Non AA camera, and will post in the next 30min?
-
hmmmm, I dont post images online, and not having much luck. Anyone have a way I can upload. I dont have any image hosting places.
-
NO, you cannot regain lost sharpening, you will only increase contrast, and preceived sharpness, no matter the channel, it is applying the same effect.
Some result better than others, but even with some samples posted here, you will see that you CANNOT regain the information that native sharpness resolved.
Phil,
As Erik already said, YES it is possible to restore blur, to a significant degree anyway. There is a lot of new research that shows even better results than from a regular Richardson Lucy (RL) restoration. You can convince yourself that RL works with a free implementation that apparently is implemented in the RawTherapee (http://www.rawtherapee.com/) converter.
Cheers,
Bart
-
Phil,
As Erik already said, YES it is possible to restore blur, to a significant degree anyway. There is a lot of new research that shows even better results than from a regular Richardson Lucy (RL) restoration. You can convince yourself that RL works with a free implementation that apparently is implemented in the RawTherapee (http://www.rawtherapee.com/) converter.
Bart
Bart,
Roger Clark (http://www.clarkvision.com/articles/image-restoration1/) has also demonstrated use of RL in lieu of or in addition to traditional unsharp masking. However, he did not explain how he derived the point spread function parameters (PSP) used for that restoration. Just guessing? I understand that deconvolution can also be used for defocusing error and and diffraction correction. Some photographers report good results using deconvolution (FocusMagic) to offset the effect of a blur filter in capture sharpening. Unfortunately, it is difficult to derive the PSP function to be used in the deconvolution. Do you have any pointers in this area?
-
Phil,
As Erik already said, YES it is possible to restore blur, to a significant degree anyway. There is a lot of new research that shows even better results than from a regular Richardson Lucy (RL) restoration. You can convince yourself that RL works with a free implementation that apparently is implemented in the RawTherapee (http://www.rawtherapee.com/) converter.
Cheers,
Bart
I am open to it, I will try it out....is this method also available to do using PhotoShop, or is this something Focus Magic specific. I will have to check out the links.
It is a bit hard for me to believe it. I almost took the chance today and purchase a 50D as it is the closest to my 14mp, so I will see within the week, as I was thinking of waiting about 3-4months and purchase the 5dMk2.....But I don't need either as I have enough bodies, and they are secondary to my MFDB, so the 50D is still something I might get...ok, enough rant.
How can I post my raw files?
-
Bart,
Roger Clark (http://www.clarkvision.com/articles/image-restoration1/) has also demonstrated use of RL in lieu of or in addition to traditional unsharp masking. However, he did not explain how he derived the point spread function parameters (PSP) used for that restoration. Just guessing?
Hi Bill,
Yes, it is basically guessing (although with visual feedback from a small image area that guess can be pretty good!), unless one has a good method of determining the characteristics of the optical system. I use Imatest to calculate the PSF of my optical system (lens/aperture/AA-filter/microlens/sensel aperture), and I am able to derive a pretty good PSF that way. The guessing part also works fine though .
I understand that deconvolution can also be used for defocusing error and and diffraction correction. Some photographers report good results using deconvolution (FocusMagic) to offset the effect of a blur filter in capture sharpening. Unfortunately, it is difficult to derive the PSP function to be used in the deconvolution. Do you have any pointers in this area?
The interesting thing is that an application like FocusMagic can already do an impressive job based on a single (radius) parameter. It is capable of tackling defocus/diffraction/AA-filtration, even motion blur, which all have distinctly different PSF shapes, so I assume its PSF assumption may be a bit more elaborate (or an average of different shapes) than a simple defocus model of a PSF, or the algorithm is somewhat adaptive. Anyway, it proves that with limited input one can already achieve impressive improvement results.
As an example, a C1 conversion of an EOS-1Ds3 image at f/8 without sharpening, and the FocusMagic version with a small PS layer blending tweak:
[attachment=19347:OPF_Debl...rop01_1_.png] [attachment=19348:OPF_Debl..._125__1_.jpg]
IMHO those who seek for a solution by removing an AA-filter will add more problems than the benefits of an AA-filtering + proper sharpening can deliver ...
Cheers,
Bart
P.S. I've kept the unsharpened image in the losslessly compressed PNG file format to allow comparisons with other sharpening utilities.
-
NO, you cannot regain lost sharpening, you will only increase contrast, and preceived sharpness, no matter the channel, it is applying the same effect.
Some result better than others, but even with some samples posted here, you will see that you CANNOT regain the information that native sharpness resolved.
As others have mentioned, sharpness can be restored if the point spread function (think of it as a lens blur profile) is known. I'm currently working on a software program that does this exact thing. All in all, my impression so far is that it is better to use an AA filter to filter out aliasing artifacts, then use deconvolution to reverse the blurring of frequencies below Nyquist.
-
Several new and exciting technologies are going to hit the imaging world and I think it is about time if we find aliasing to be too objectionable to just use a camera which has an AA filter and move on.
Care to share what those might be?
-
Care to share what those might be?
Yes, for one thing: Hyper-/multi-imaging. I have talked about it a little here on this forum before. We have worked a little on that and are blown away at the prospects of it - ranging all the way from food inspection, to digital special effects especially for filmmaking, to medical imaging, to early-cancer detection, to user-specific camera curves resulting in more control than traditional 3-color RGB, to digital pathology, to ....
However, I'm not sure if Adobe/Apple/etc. are yet ready for it. New algorithms might need to be developed for such data.
Roger Clark (http://www.clarkvision.com/articles/image-restoration1/) has also demonstrated use of RL in lieu of or in addition to traditional unsharp masking. However, he did not explain how he derived the point spread function parameters (PSP) used for that restoration. Just guessing? I understand that deconvolution can also be used for defocusing error and and diffraction correction.
Those looking for a general solution might google "blind image deconvolution" for which you may not need to know the psf. However, in theory that is fine, but in practise many constraints are imposed such as the Gaussian assumption that helps in closed form solutions for methods such as maximum-likelihood. Richardson-Lucy algorithm can be shown to be a special case of these general frameworks.
-
As others have mentioned, sharpness can be restored if the point spread function (think of it as a lens blur profile) is known. I'm currently working on a software program that does this exact thing. All in all, my impression so far is that it is better to use an AA filter to filter out aliasing artifacts, then use deconvolution to reverse the blurring of frequencies below Nyquist.
Indeed, I do as well, and it's my impression that the vast majority of the Digital Signal Processing (DSP) community is of the same opinion. When over-sampling is impractical, add a low-pass filter before sampling and then restore.
Cheers,
Bart
-
Those looking for a general solution might google "blind image deconvolution" for which you may not need to know the psf. However, in theory that is fine, but in practise many constraints are imposed such as the Gaussian assumption that helps in closed form solutions for methods such as maximum-likelihood.
The main issue with blind deconvolution methods is noise amplification.
Richardson-Lucy algorithm can be shown to be a special case of these general frameworks.
In a way, yes because RL is based on a Bayesian approach to Poisson noise (just like Photon shot noise), but with improvements like additional insensitivity to random noise.
Cheers,
Bart
-
As others have mentioned, sharpness can be restored if the point spread function (think of it as a lens blur profile) is known. I'm currently working on a software program that does this exact thing. All in all, my impression so far is that it is better to use an AA filter to filter out aliasing artifacts, then use deconvolution to reverse the blurring of frequencies below Nyquist.
Maybe I can email the file to someone who can post it.
-
Hi,
Many of us have used:
http://www.yousendit.com/ (http://www.yousendit.com/)
For sending pictures.
You can mail to yourself. Yousendit sends a short mail with a download link. You can just post the download link on this forum.
Maybe I can email the file to someone who can post it.
-
Hi,
RawDeveloper from Iridient has also Richardson Lucy, in part probably because RL is part of the image processing libraries on Apple. I presume that it's based on a guess regarding PSF.
My guess is that it may be possible to guestimate PSF in some cases. The AA-filter acts a beam splitter so it probable deflects some of the light in four adjacent cells.
http://books.google.com/books?vid=ISBN0240...2yfoNY#PPA30,M1 (http://books.google.com/books?vid=ISBN0240515900&id=wsxk03-gceUC&pg=PA30&lpg=PA30&ots=BlxgR6ZXmz&dq=anti-aliasing+lithium-niobate&sig=YUsQZMtxjtq70oOYWmU1K2yfoNY#PPA30,M1)
Best regards
Erik
Bart,
Roger Clark (http://www.clarkvision.com/articles/image-restoration1/) has also demonstrated use of RL in lieu of or in addition to traditional unsharp masking. However, he did not explain how he derived the point spread function parameters (PSP) used for that restoration. Just guessing? I understand that deconvolution can also be used for defocusing error and and diffraction correction. Some photographers report good results using deconvolution (FocusMagic) to offset the effect of a blur filter in capture sharpening. Unfortunately, it is difficult to derive the PSP function to be used in the deconvolution. Do you have any pointers in this area?
-
Hi,
Many of us have used:
http://www.yousendit.com/ (http://www.yousendit.com/)
http://www.yousendit.com/download/VGlmbUpRNDRHa04zZUE9PQ (http://www.yousendit.com/download/VGlmbUpRNDRHa04zZUE9PQ)
-
http://www.yousendit.com/download/VGlmbUpRNDRHa04zZUE9PQ (http://www.yousendit.com/download/VGlmbUpRNDRHa04zZUE9PQ)
This is the longest pause this post has had...anyone DL the image and look it over?
-
http://www.yousendit.com/download/VGlmbUpRNDRHa04zZUE9PQ (http://www.yousendit.com/download/VGlmbUpRNDRHa04zZUE9PQ)
I just opened this image up to expose it a bit, and I lowered the color noise some also, and I hardly added some sharpening, and it is rather tack without any sharpening.
-
Hi,
I have downloaded and had a look. Don't know what to say...
I don't see artifacts jumping at me.
Best regards
Erik
This is the longest pause this post has had...anyone DL the image and look it over?
-
Hi,
I have downloaded and had a look. Don't know what to say...
I don't see artifacts jumping at me.
Best regards
Erik
Perhaps they are not jumping...Are there any artifacts like we have been discussing?..maybe point a couple things out?
-
I converted it to DNG. Here it is for anyone else who wants it for convenience: FM9R3347.dng (http://thebrownings.name/photo/2010-01-aliasing/FM9R3347.dng).
...anyone DL the image and look it over?
I have a few observations. Most of the image is outside the plane of critical focus, including the entire leaf and much of the ruler. The thin DOF (around one inch in this photo) precludes aliasing. The sharpness is not as high as I would expect for an aliasing test. None of the subjects provide sufficiently fine details (at high contrast). If the ruler was several feet away, then the numbers, letters, and lines would be small enough to serve that purpose. As it is, the smallest numbers are over 20 pixels wide. Only the scuff marks would be small enough to demonstrate aliasing, but their contrast is too low to do it. It seems to me that the only area that could alias is the table itself (the part of it that is in critical focus), since it is the only part of the image with fine lines. And I do indeed see some aliasing there: some of the lines have jaggies. I see what could be some terrible chroma aliasing, but given the age of this sensor, I bet it's just noise.
-
Perhaps they are not jumping...Are there any artifacts like we have been discussing?..maybe point a couple things out?
Hi Phil,
The DOF seems to be shallow, and the focus is maybe at a distance where the lens is not at it's best. Using out-of focus areas functions as an effective low-pass filter. The aperture data is missing from the EXIF, which makes it difficult to come to a conclusion, other than that this is not a shooting scenario that will usually show a lot of aliasing.
It also doesn't look like something that's better than could be expected from a camera with an AA-filter + proper sharpening, so for me it is not a useful demonstration of ..., whatever it should demonstrate.
Cheers,
Bart
-
The #16 on the ruler is what is focused on. This Should have shown defraction as it is F22. I took also at two other stops.
Maybe I should back out another foot?
-
Hi,
f/22 would act as a very crude AA-filter.
Check this figure: http://83.177.178.7/ekr/images/stories/difractionlimit.gif (http://83.177.178.7/ekr/images/stories/difractionlimit.gif)
Best regards
Erik
The #16 on the ruler is what is focused on. This Should have shown defraction as it is F22. I took also at two other stops.
Maybe I should back out another foot?
-
Hi,
f/22 would act as a very crude AA-filter.
Check this figure: http://83.177.178.7/ekr/images/stories/difractionlimit.gif (http://83.177.178.7/ekr/images/stories/difractionlimit.gif)
Best regards
Erik
Do you think there would be a difference for the sharpness in the focus area at F11 or 8?
The relation of this lens and the sensor, are rather very nice, as you don't really see defraction. At least I don't.
-
Do you think there would be a difference for the sharpness in the focus area at F11 or 8?
Green wavelengths at anything narrower than f/8, maybe f/9, will probably be visibly affected by diffraction due to the diffraction pattern exceeding the sensel size of the Kodak by a significant margin. How the lens and effective sensel aperture interact, I don't know.
Cheers,
Bart
-
Green wavelengths at anything narrower than f/8, maybe f/9, will probably be visibly affected by diffraction due to the diffraction pattern exceeding the sensel size of the Kodak by a significant margin. How the lens and effective sensel aperture interact, I don't know.
Cheers,
Bart
I have photographed the same image in 8 and 11 and 16.
I would like to learn more, but my observations when I tested a Canon vs this Kodak, at a given aperture, the files are MUCH crisper, much sharper, much more 3d like. I do remember when I ran the test that the Canon 1Ds(11mp) would show deffraction before the Kodak(I could be wrong and I will test it as soon as I get all my Canon bodies back).
If I am able to remove all of the Canon AA filter, I would. But the fact that only 1 is removalable, makes it little point. The Sony A900 is another option. If it is able to be AAfilter free, I would not hesitate to take the Sony over the Canon, simply for this reason.
These to me are tools, and may the most adaptable tool be the one I use for the right job.
-
Hi,
The effect is called diffraction. Diffraction is simply laws of physics, nothing lens or sensor can do about it. In general diffraction starts to deteriorate an image around or past f/8, depending on
- residual errors in lens
- wavelength of light
There is no way of counter acting diffraction. The only way deterioration by diffraction can be avoided at small apertures is bad technique:
- A really awful lens
- Bad focus
- Camera vibrations
The diffraction will still be there but it may be masked by the other effects. On the positive side, diffraction will reduce/eliminate aliasing and it responds well to simple sharpening methods like unsharp mask.
Here is a good discussion of diffraction: http://www.cambridgeincolour.com/tutorials...photography.htm (http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm)
That said, the human vision can be quite tolerant of unsharpness. I once had a discussion about the effect of MLU on tripod, so I shot a series of test shots with/without MLU. Both series of images looked sharp to me, but when I actually measured MTF with Imatest half the resolution was gone when not using MLU, so I really converted my 12.5 MP camera into a 3 MP camera. Rechecking the images I could see that there was a loss of sharpness. Often you see what you expect/hope to see.
Best regards
Erik
Do you think there would be a difference for the sharpness in the focus area at F11 or 8?
The relation of this lens and the sensor, are rather very nice, as you don't really see defraction. At least I don't.
-
Hi,
The effect is called diffraction. Diffraction is simply laws of physics, nothing lens or sensor can do about it. In general diffraction starts to deteriorate an image around or past f/8, depending on
- residual errors in lens
- wavelength of light
There is no way of counter acting diffraction. The only way deterioration by diffraction can be avoided at small apertures is bad technique:
- A really awful lens
- Bad focus
- Camera vibrations
The diffraction will still be there but it may be masked by the other effects. On the positive side, diffraction will reduce/eliminate aliasing and it responds well to simple sharpening methods like unsharp mask.
Here is a good discussion of diffraction: http://www.cambridgeincolour.com/tutorials...photography.htm (http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm)
That said, the human vision can be quite tolerant of unsharpness. I once had a discussion about the effect of MLU on tripod, so I shot a series of test shots with/without MLU. Both series of images looked sharp to me, but when I actually measured MTF with Imatest half the resolution was gone when not using MLU, so I really converted my 12.5 MP camera into a 3 MP camera. Rechecking the images I could see that there was a loss of sharpness. Often you see what you expect/hope to see.
Best regards
Erik
Deffraction is the result of (as I unsderstood it) is the reaction of light off the aperture blades. the closer they get to each other and the relation to the size of the hole. Varing lens will deffract differntly depending on the size/cut/shape of the aperture blades.
-
Hi,
You are right, sort of, but the effect you are thinking about is second order when you stop down long enough, like beyond f/8. The cause of diffraction is the size of the hole not the blades themselves. I'd suggest that you effect you are describing is probably for real, but not relevant in this context. I'd suggest that you check the link I had in the previous posting, or any textbook.
Best regards
Erik
Deffraction is the result of (as I unsderstood it) is the reaction of light off the aperture blades. the closer they get to each other and the relation to the size of the hole. Varing lens will deffract differntly depending on the size/cut/shape of the aperture blades.
-
Hi,
You are right, sort of, but the effect you are thinking about is second order when you stop down long enough, like beyond f/8. The cause of diffraction is the size of the hole not the blades themselves. I'd suggest that you effect you are describing is probably for real, but not relevant in this context. I'd suggest that you check the link I had in the previous posting, or any textbook.
Best regards
Erik
I have tested a leica 100 vs canon 180 canon. The Canon lens itself is very sharp, but when stopped, The canon shows deffraction sooner, the Leica performed better f22 vs f22. This is from memory, I will have to revisit.
Learning these things is one thing. My intention is to making a tool... Process of elimination is my approach of learning for choosing the right tool :-)
-
In a way, yes because RL is based on a Bayesian approach to Poisson noise (just like Photon shot noise), but with improvements like additional insensitivity to random noise.
Richardson-Lucy converges to maximum likelihood solution. The maximum likelihood solution may not be the "best" solution in this case but it is unbiased. Of course, the whole thing may be seen in light of Bayesian statistics.
-
I have tested a leica 100 vs canon 180 canon. The Canon lens itself is very sharp, but when stopped, The canon shows deffraction sooner, the Leica performed better f22 vs f22. This is from memory, I will have to revisit.
Learning these things is one thing. My intention is to making a tool... Process of elimination is my approach of learning for choosing the right tool :-)
perhaps, the relation of the lens to the sensor pixels Im sure play a larger part
-
After much thinking if I should get the 50D to do the test, I figured there would be yet another crop facor. So the closest one I thought would make sense also keeping the camera was the 5Dii, so I purchased it. It doesn't have a 50 ISO setting. but from looking at the files...this is a rather nice camera.
My set did move from the time I posted the image to now, so i will do an apple/apple image comparison. and do the YouSend...but from the last post, Everyone loved explaining their own understanding , and looked like there were plenty to want to listen or learn, but no one but Erik cared to actually look at the test, or the first image of at least...your thoughts?
-
Hi,
To begin with I hope that you like your new camera!
The other issue I would say that you kicked of several good discussions and several knowledgeable persons chimed in. It may even be that we get some new and better tools because new ideas have been discussed.
Best regards
Erik
After much thinking if I should get the 50D to do the test, I figured there would be yet another crop facor. So the closest one I thought would make sense also keeping the camera was the 5Dii, so I purchased it. It doesn't have a 50 ISO setting. but from looking at the files...this is a rather nice camera.
My set did move from the time I posted the image to now, so i will do an apple/apple image comparison. and do the YouSend...but from the last post, Everyone loved explaining their own understanding , and looked like there were plenty to want to listen or learn, but no one but Erik cared to actually look at the test, or the first image of at least...your thoughts?
-
Daniel why is the frequency of the moire in the non AA image on the rivets for the diagonal beam low and if indeed each rivet is represented by more than one pixel why are some rivets missing, unless of course, these rivets are not bright chrome but colored and/or missing. Is it not a possibility that the AA filter reconstructed missing rivets from surrounding pixels by averaging; are all the rivets exactly uniform?
Excellent! Most people who see this picture prefer the aliased version, so it's nice to meet another person who dislikes the aliasing in it as much as I do, at least for images where the aliasing is extremely bad.
Now the only question is about circumstances where the aliasing is much more subtle (which is usually the case when comparing filtered-vs-filterless cameras).
I am talking about raw. The reason I gave you the extreme example was to set a baseline. If you had preferred the aliased image in the extreme example (as many do), then I would know that you prefer the aliased image in a more subtle example for sure.
Now let's take a look at the following somewhat flawed comparison (http://forums.dpreview.com/forums/read.asp?forum=1019&message=30112382) between an SD14 (no AA filter) and a 50D (relatively weak AA filter).
Here is the SD14 100% crop:
(http://thebrownings.name/photo/2010-01-aliasing/106242475.k6guNjIC.SDIM3020-crop.png)
And here is the 50D 100% crop (after downsampling to the same pixel count):
(http://thebrownings.name/photo/2010-01-aliasing/IMG_3026-rawtherapee-4mp-crop.png)
The full SD14 image (http://thebrownings.name/photo/2010-01-aliasing/106242475.k6guNjIC.SDIM3020.jpg)
The full 50D image. (http://thebrownings.name/photo/2010-01-aliasing/IMG_3026-rawtherapee-4mp.png)
The purpose of the comparison was not to show the difference between filtered and unfiltered cameras, so it is somewhat flawed for the purpose of this thread, but I think it can still illustrate some of the types of aliasing artifacts that are seen with filterless cameras.
This example does not have aliasing that is quite as extreme as my last example, but it is a man-made subject, where more people tend to object to the aliasing artifacts.
For example, there appear to be many missing rivets in the SD14 image, while they are all there in the 50D image.
Do you feel the same way about this aliased image as you did the last (i.e. "horrible")? Or do you see the aliased image as having better sharpness/3D than the anti-aliased one? If the former, then the next task will be to find a comparison where the difference in aliasing is more subtle.
-
I recommend then to all of AA's troubled to purchase a SD14! if the AA is disturbing your thoughts. The Sigma is just down-priced to such ridiculous level that ones can buy it just to try.
I've worked with the SD14 during 4 years, both for pro and amateur works.
You know what? Just try it, but in any kind of situations, and you will see rapidly by yourself the...AA.
Fred
-
I recommend then to all of AA's troubled to purchase a SD14! if the AA is disturbing your thoughts. The Sigma is just down-priced to such ridiculous level that ones can buy it just to try.
I've worked with the SD14 during 4 years, both for pro and amateur works.
You know what? Just try it, but in any kind of situations, and you will see rapidly by yourself the...AA.
Fred
I have tried without the AA... using the SLRC Kodak, and that is why I am a "Pro Optional AA filter advocate" :-)
Just as I did this test when all these posts were going on, I was about to purchase a 50D to do the test, and then decided to get the 5Dmrk2, and It is a very nice camera, and I really like the files. I will have to get some apple to apple files to show sometime soon.
-
Hi Joe, thanks for the response.
Daniel why is the frequency of the moire in the non AA image on the rivets for the diagonal beam low?
If you're saying that you see little or no aliasing (moire) of the rivets on the diagonal beam of the non AA image, then I disagree; to me it is aliasing there as well.
If you're asking why the spatial frequency at which the aliasing of the rivets occurs is so low, I would say that it's actually not very low at all -- in fact it's quite close to Nyquist. Really bad moire occurs at much lower spatial frequencies.
if indeed each rivet is represented by more than one pixel why are some rivets missing, unless of course, these rivets are not bright chrome but colored and/or missing.
Because when you practice unsafe imaging (without a protective AA filter), you contract certain artifacts: whether or not you see the rivet depends on chance alignment of the pixel grid.
Is it not a possibility that the AA filter reconstructed missing rivets from surrounding pixels by averaging;
No. The Bayer demosaic algorithm does use a dozen or more of the surrounding pixels to interpolate color values, but that occurs at a much higher level of detail than what you're seeing in this image. Remember that this is only 4 MP -- the original is 15 MP.
are all the rivets exactly uniform?
Download the original 15 MP raw file if you want to see what they really look like.
-
Because when you practice unsafe imaging (without a protective AA filter), you contract certain artifacts: whether or not you see the rivet depends on chance alignment of the pixel grid.
Daniel,
As Joe was asking, if the rivets are larger than a single pixel, as they appear to be, how could they "fall between the cracks" and disappear due to misalignment with the pixel grid?
It seems to me that some of the rivets do not appear because of subtle differences in the direction of the glancing light. It is just as easy to pick out areas where rivets appear in the Sigma shot but not in the Canon shot.
Regards,
Cliff
-
To know for sure how much aliasing is present on the rivets, you would need two shots with the camera displaced slightly. The form of the alias (as constructive or destructive of a small detail) does depend on the relative position of the detail and the pixel structure. This is much more obvious in a motion imaging device if the camera moves slightly.
The first devices to invoke serious consideration of aliasing were standard definition television cameras, because of the relatively coarse scanning line structure, equivalent to 480 pixels per picture height. This was so coarse that you could never escape a compromise between small detail contrast (vertical frequency response and sharpness) and aliasing. In tube cameras, the aliasing was relieved by making the scanning spot blurry enough so that the scanning lines essentially overlapped their neighbors. In addition to the simple spatial aliasing, there was worse aliasing due to interlaced scanning, also relieved by the blurriness of the scanning spot. When TV cameras went from tubes to CCDs, the CCDs had to be designed to average two adjacent rows of pixels for each interlaced field in order to reduce aliasing to an acceptable level. Then came high definition. With about 1000 rlines (rows of pixels), the aliasing was much less objectioanble, parlty because the lens was now a prefilter for antialiasing. The CCD now could have a variable amount of averaging between adjacent rows to give a choice of more detail contrast (but with some aliasing) or less detail contrast (but cleaner). NOTE that this is still a motion imaging case, so you have to go more towards clean than towards detail contrast, since whenever the camera moves, the variation in the alias components may be objectionably strong.
For digital still cameras, since you do not have the temporal variation in the aliasing making it so obvious, you may get away with a much higher level of aliasing, which gives greater sharpness, albeit with "false" detail. It's a matter of taste. My personal taste for a sensor with the high resolution of a DSLR is to keep the antialising filter and boost the detail contrast with post sharpening.
On the other hand, in designing consumer digital converter boxes to go from high-definition broadcasts to analog TV sets, it has been found that a filter that completely eliminates aliasing looks far too soft - it's back to the same trade-offs as for a standard definition TV camera - some detail and some aliasing.
Also by the way, there is a similar principle in loudspeaker design - a mid-range driver with a small amount of distortion will make a speaker sound like it has better high frequency response than it really does - but the high frequencies are being generated in the speaker and didn't really come from the source. Depending on how much distortion is present, some sources may not sound their best.
-
Hi,
I downloaded both images a while ago, but at that time I couldn't find the missing rivets, simply because I looked at the wrong place. Now I can see the effect clearly. The problem with this comparison is that there is no way around the fact that the D50 resolves much higher the the Sigma SD-14. We could downscale the Canon image to the same size as the SD-14.
I tried it using Photoshop "bicubic" and made the following observations:
The rivets are still there in the Canon image.
The next try I made was resize both images 200%, mostly for better viewing. The Canon image scales better. There are lots of jaggies on diagonals in the Sd-14 image while the Canon image is still pretty smooth.
So my conclusion is that the resolution advantage of the Canon is still there when the image is downscaled to 5 MPIxels. Rescaling the 5 MP images to 20 MP (200%) works better for Canon 50D. The SD-14 shows clearly more jaggies.
I don't enclose the images, as I have a problem with my Internet Provider and need to use low bandwidth solution, but anyone can retry with the original images.
Best regards
Erik
Hi Joe, thanks for the response.
If you're saying that you see little or no aliasing (moire) of the rivets on the diagonal beam of the non AA image, then I disagree; to me it is aliasing there as well.
If you're asking why the spatial frequency at which the aliasing of the rivets occurs is so low, I would say that it's actually not very low at all -- in fact it's quite close to Nyquist. Really bad moire occurs at much lower spatial frequencies.
Because when you practice unsafe imaging (without a protective AA filter), you contract certain artifacts: whether or not you see the rivet depends on chance alignment of the pixel grid.
No. The Bayer demosaic algorithm does use a dozen or more of the surrounding pixels to interpolate color values, but that occurs at a much higher level of detail than what you're seeing in this image. Remember that this is only 4 MP -- the original is 15 MP.
Download the original 15 MP raw file if you want to see what they really look like.
-
The next try I made was resize both images 200%, mostly for better viewing. The Canon image scales better. There are lots of jaggies on diagonals in the Sd-14 image while the Canon image is still pretty smooth.
So my conclusion is that the resolution advantage of the Canon is still there when the image is downscaled to 5 MPIxels. Rescaling the 5 MP images to 20 MP (200%) works better for Canon 50D. The SD-14 shows clearly more jaggies.
Eric,
In contrast to the 50D file, the provided SD14 jpeg has strong sharpening halos. If you enlarge that jpeg you will also enlarge the jaggie sharpening artifacts.
I made a comparison in another thread (http://luminous-landscape.com/forum/index.php?showtopic=40809&view=findpost&p=342105) by going back to the raw files and not sharpening either image. As expected the Canon is still better but the differences are not so glaring - or what do you think?
Cliff
-
Lets see some SD1 Files!!!
-
Lets see some SD1 Files!!!
I'll show you mine if you show yours first ;)
Cheers,
Bart
P.S. Truth in advertizing forces me to confess I don't have any