Luminous Landscape Forum
Equipment & Techniques => Cameras, Lenses and Shooting gear => Topic started by: Kenneth Sky on March 08, 2007, 02:08:15 pm
-
It seems that Sigma has listened to the feedback on announcing the advent of the DP-1 at Photokina: external optical viewfinder, hotshoe for flash and RAW capability. I'll bet it becomes the poor man's Leica M8
-
f4 lens, sorry sigma, if it had been a 35mm f2.0 I would have bought it...
-
I'm curious about the crop factor. At 1.5x a 16.6mm lens equates to 24.9mm. That's closer to 24mm than to 28mm.
I also would have liked a faster lens, but I have to see what the high ISO performance looks like before making a decision.
-
The DP-1 will have a crop factor of 1.7, it uses the same sensor as in the SD14.
I wont start arguing about whether it will beat the M8 IQ-wise or not as there are no samples available so far and the cameras have different concepts (RF vs Compact) so its more or less a pointless comparison but..there are some pretty outstanding Sd14 samples around by now.
I am curious what Mr. Reichmann thinks about the new Sigmas and hope he will get his hands on one.
I will most definitely get my hands on one as soon as it is being released.
O.
-
Apparently the limitation of size and the need to cover all of the larger than normal compact sensor made an f2.8 lens impossible. As stated above if the higher ISO performance is good then that may compensate.
-
f4 lens, sorry sigma, if it had been a 35mm f2.0 I would have bought it...
[a href=\"index.php?act=findpost&pid=105507\"][{POST_SNAPBACK}][/a]
f/2.0 would be the way to go. At least this is a start.
I'll be interested to see how this generation of the X3 chip performs. Previous chips had issues with green blotchy noise. If they fixed that, then this will be something to look forward to.
On a side note, I wish they would stop with this x3 resolution marketing crap. It's a 4.7MP sensor. Not 14. Yes, it resolves more than a bayer pattern sensor but it's not 3 times as much. They'd be better off making a sensor that actually is about 10MP in pixel resolution and then advertising they resolve more detail than any of the competition, even in the high-end of things. I don't understand why they keep aiming lower than what they could do.
At the very least news sources need to stop the heck out of reporting the marketing as fact.
-
At the very least news sources need to stop the heck out of reporting the marketing as fact.
[a href=\"index.php?act=findpost&pid=105564\"][{POST_SNAPBACK}][/a]
I cringed when I saw the B&H website list the SD10 as 10.3MP.
I wonder if there is a way to get a mild AA filter onto this thing. Maybe I can attach a vibrator to it. I'd like one for B&W; ironically, *that* is what the Foveon excels at. The noise levels are very low in the Foveon before a converter extrapolates color from it. An ISO 800 pushed a few stops stills looks fairly clean in B&W.
-
This camera looks very interesting... it's a shame the lens is only f/4, but given the sensor size you can just up the ISO and it should still ace a point-and-shoot with f/2.8 but a tiny sensor. This seems a perfect compact camera for when lugging around my 5D isn't appropriate, I just wonder what it's going to cost...?
-
Yes, it needs an anti-alias filter. Every foveon image I've ever seen has been full of jaggie artifacts that I know sigma-crazies love, but that's not the way my eyes see the world. Without an anti-alias filter, I'd say the camera is incorrectly engineered. At least with all the bayer sensor cameras, you've got enough resolution you can downscale and get increased per-pixel sharpness without aliassing, but with the sigmas, you're starting off with such a low per-pixel resolution you're pretty much stuck with what you've got. Then, there's the hyper-strong matrix needed to extract proper colour from the camera RGB space tbat ruins you're noise performance. Given the size of the pixels it should be a high-ISO, low noise monster, but it isn't. What you're left with is a paradox of a camera. Interesting, but that's it....
Graeme
-
It seems that Sigma has listened to the feedback on announcing the advent of the DP-1 at Photokina: external optical viewfinder, hotshoe for flash and RAW capability. I'll bet it becomes the poor man's Leica M8
[a href=\"index.php?act=findpost&pid=105498\"][{POST_SNAPBACK}][/a]
What is the lead time in the deign, production and marketing of a new camera? The question is, how long ago did Sigma have to start listening?
-
The problem with this specialized a product is that everyone (almost) will have at least one show stopper. For me, it's not the f4, but rather the lack of articulated screen. I'd use this for unobtrusive street shooting, but after the Sony R1 I'll never give up an articulated LCD. Just like I won't settle for no raw or more noise.
-
Yes, it needs an anti-alias filter. Every foveon image I've ever seen has been full of jaggie artifacts that I know sigma-crazies love, but that's not the way my eyes see the world.
Thats because the images may have been processed wrongly. Its in part due to a misleading setting of the Raw Converter. Sharpening needs to be set to -0.8 to switch any sharpening off. You can imagine what artifacts are introduced if the image is sharpened in the first place (e.g. with fringing and such things not removed beforehand), processed in PS with curves or anything, then up- or downsized and then sharpened again.
I usually forget to switch off sharpening in the Raw converter and it shows in my online images.
Did you look at the SD14 sample images? I see no jaggies in there.
Without an anti-alias filter, I'd say the camera is incorrectly engineered. At least with all the bayer sensor cameras, you've got enough resolution you can downscale and get increased per-pixel sharpness without aliassing, but with the sigmas, you're starting off with such a low per-pixel resolution you're pretty much stuck with what you've got.
Disagree but dont want to get into an argument "Bayer vs Foveon" here..
But I was wondering what you meant with per-pixel resolution. ANd I think I am safe to say that Foveon images size up better than bayer images with a similar amount of pixels in them.
Then, there's the hyper-strong matrix needed to extract proper colour from the camera RGB space tbat ruins you're noise performance. Given the size of the pixels it should be a high-ISO, low noise monster, but it isn't. What you're left with is a paradox of a camera. Interesting, but that's it....
Graeme
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=105638\")
Agree thats a weak point. There is a software which extracts the colors as recorded by the camera. The colors are "off" (yet in a very pleasing way for me) but the tonal nuances captured are amazing (speaking of the SD10 here). Example here:
[a href=\"http://www.flickr.com/photo_zoom.gne?id=407216923&size=l]http://www.flickr.com/photo_zoom.gne?id=407216923&size=l[/url]
O.
-
I see aliassing in pretty much every sigma image I've ever seen. Which is a shame as it's such an easy thing to have removed by a small filter infront of the sensor. But, because Sigma has such lower resolution than it's competition, they don't put this necessary filter in place as it does reduce resolution a little.
I guess some people are more annoyed by aliasing artifacts than others, depending on how their visual system is wired up. Also, when you know the engineering and mathematics of sampling theory, you associate such aliassing with an artifact, rather than the fake details that perhaps you'd think it was otherwise. You could view the aliassing as either an artifact or fake detail, but it's the brain that colours your perception, and colours mine to think it looks bad.
-
Actually the microlenses of the SD10 work as a very light AA filter.
Did you look at the fullsize sample images on the Sigma homepage? Do you see aliasing in those images as well?
-
I agree there can be an aliasing issue with the Foveon images. It shows in the sample raw images I have and seen but it is very dependent on the angle of the edge in question. Some images it shows quite a bit and some it does not. The solution is not to add an AA filter, but to increase the resolution of the sensor.
Ollivr, you say Foveon sensors can uprez better than other sensors of similar resolution. I would agree with that but there is a problem. When everyone was selling 6MP sensors, Sigma had a 3.4MP sensor. Now that everyone is at 10MP for APS-C, Sigma has moved up a whopping 1.2MP to 4.6MP (and it only took them 3 years to get there!). All your ability to uprez a Foveon file will go to simply matching what you can get from the start out of a Canon or Nikon in the same market segment.
Despite this, I think the Foveon chip will work will better in the form factor of the DP-1 than a SLR. With a SLR people are expecting high-performance imaging performance that reaches the bar Canon has set in the past. However, with this unique form-factor it is a specialty camera that is best suited for photography that typically is not printed large. And so the anemic resolution becomes nearly a non-issue.
The things that really are worth wondering, IMHO, about with this camera are ISO performance, lens performance, and responsiveness.
-
Ollivr, you say Foveon sensors can uprez better than other sensors of similar resolution. I would agree with that but there is a problem. When everyone was selling 6MP sensors, Sigma had a 3.4MP sensor. Now that everyone is at 10MP for APS-C, Sigma has moved up a whopping 1.2MP to 4.6MP (and it only took them 3 years to get there!). All your ability to uprez a Foveon file will go to simply matching what you can get from the start out of a Canon or Nikon in the same market segment.
I put the comparison in that a conservative way in order to avoid a "Bayer vs Foveon" thing here. I couldnt care less what camera or sensor somebody uses, although I have to admit that I personally prefer the Foveon from my limited experience. Still, I found the opinions about foveon technology a little unfair (nobody notices the artifacts bayer cams have?) which is why I replied to this thread.
In some weeks there may be reviews saying the SD14 matches a typical 10MP bayer or that it doesnt, and people will trust those tests or not. I certainly neither have the equipment nor the experience needed to conduct a proper test. I can only look at Bayer images, and Foveon images (made from cameras of similar age and price range), and draw an unscientifical conclusion for myself.
Regarding hi ISO: Dont expect miracles there from Sigma, it has certainly improved but is not up to bayer standards. One should think of this whole foveon thing rather as an alternative type of film. Different behavior, different advantages and disadvantages.
Regards,
O.
-
On a side note, I wish they would stop with this x3 resolution marketing crap. It's a 4.7MP sensor. Not 14. Yes, it resolves more than a bayer pattern sensor but it's not 3 times as much. ....
At the very least news sources need to stop the heck out of reporting the marketing as fact.
[a href=\"index.php?act=findpost&pid=105564\"][{POST_SNAPBACK}][/a]
A little OT but I'd have to say that it is really only their timing that is off !
It was the original digital cameras with Bayer arrays that abused the meaning of the word pixel but no-one grumled about that marketing cr*p When 640x480 color monitors came out they still only had ~300,000 pixels and each pixel was assumed to be complete (i.e. RGB) and they were not described as 920,000 pixel monitors. Even the hype about "full" HDTV ( i.e. 1080p60) claims only 2 million pixels when they could claim 6
They'd be better off making a sensor that actually is about 10MP in pixel resolution and then advertising they resolve more detail than any of the competition, even in the high-end of things. I don't understand why they keep aiming lower than what they could do.
[a href=\"index.php?act=findpost&pid=105564\"][{POST_SNAPBACK}][/a]
My guess, err, is that these sensors are kinda difficult to make
-
Actually the microlenses of the SD10 work as a very light AA filter.
A microlens may reduce the intensity of aliasing a little bit, but it is not an AA filter per se. In order to have proper AA filtering, a photosite has to capture some photons that actually struck the sensor sandwich closer to the center of another, adjacent pixel. A microlens does not do that.
Did you look at the fullsize sample images on the Sigma homepage? Do you see aliasing in those images as well?
[a href=\"index.php?act=findpost&pid=105863\"][{POST_SNAPBACK}][/a]
I saw them, and some were extremely aliased, disturbingly so.
I don't know how anyone can claim that Sigma digital images are printable at large sizes. They look like hard mosaics if printed literally, and ramped, softened mosaics, if filtered. They scream, "look at me; I'm a grid of aliased pixels", IMO.
Look at the architecture shots, the one with the row of red-brick buildings. The edges of the bricks on the corners are all exaggeratedly in the wrong place because of the aliasing.
-
Look at the architecture shots, the one with the row of red-brick buildings. The edges of the bricks on the corners are all exaggeratedly in the wrong place because of the aliasing.
[a href=\"index.php?act=findpost&pid=106032\"][{POST_SNAPBACK}][/a]
I see how the bricks look weird at the edges but I wonder wether it is camera body artifacts. Just doesnt look like aliasing to me. As far as the jaggies are concerned I can only tell you about my own: Its mostly user error due to not setting sharpening to -0.8 in the raw converter.
O.
-
Sharpening just makes the inherent aliassing more visible, and if that's an issue, the Sigma that designed their raw conversion software sound like they've not spoken to the Sigma that designed the cameras. RAW conversion and camera have to work as a team to "get it right". Even the Sigma adverts in magazines show nasty aliassing on the girl's eyelashes.
Aliassing on, say, perfect computer generated images, looks like your traditional stair stepping, but because real life has lots of detail that isn't in straight lines, aliassing from that manifests as almost a noise where you're getting uncorrelated pixels next to each other as neighbouring pixels are being fed widely different colour or tonal values. In computer graphics, they vastly super-sample the image to get a smooth, continous image that's not aliassed. To do that on a sigma, would require so many more pixels than it has that the size of the camera would be like a truck. Or you could slip in a small sliver of glass, an anti-alias filter which really is an engineering necessity, take the resolution hit, but get smooth, continous images.
Graeme
-
Well anyways although I am a quite happy user of a Foveon-based camera I dont want to advertise/defend their product any more than this (except for they pay me for it). Let me just say its a different way of capturing images with different artifacts. Its a relatively young technology (more so than the bayer grid) and still has some disadvantages, one of which being that it is less user-error-proof. Advantages are the sharpness and rendition of subtle color transitions (if you do it right). Relative to the small community of foveon users, there are loads of bad images out there as you just cannot really "point and shoot" the older Sigma cameras. Lets see what MR thinks about the newer Sigmas.
O.
-
It seems that Sigma has listened to the feedback on announcing the advent of the DP-1 at Photokina: external optical viewfinder, hotshoe for flash and RAW capability. I'll bet it becomes the poor man's Leica M8
[a href=\"index.php?act=findpost&pid=105498\"][{POST_SNAPBACK}][/a]
Indeed, Sigma does listen to the feedback that they get from users. I had the opportunity to handle the DP1---it's a very sweet little camera. The accessory viewfinder is very nice. As mentioned previously, the f4 lens takes up quite a bit of space already---I doubt they can go with f2.8 without making the body taller. Clever manual focusing mechanism as well
With the new imaging pipeline for this product (TRUE), I believe the hope is for a stop or two better than the SD14. I got to shoot the SD14 for a few days, and it produces outstanding images. I got to see some head to head comparison of "artifacts" from the SD14 vs CFA sensor cameras (think D200, 5D). While there can be some luminence aliasing in the SD14, it's way better than the trashing that goes on with regards to color moire (and it's removal) in the CFA cameras. The edge aliasing is really not an issue when printing 99% of Sigma photos---and the prints at PMA were certainly proof of that.
It's a very exciting introduction---I expect it will do very well with photographers.
-
JLK
thanks for your very interesting report of the DP1 and the SD14.
Did Sigma give you any hints of shipping dates of the DP1? Cant wait to get my hands on one.
-
JLK
thanks for your very interesting report of the DP1 and the SD14.
Did Sigma give you any hints of shipping dates of the DP1? Cant wait to get my hands on one.
[a href=\"index.php?act=findpost&pid=106383\"][{POST_SNAPBACK}][/a]
I have heard "hints", but I hesitate to put stock in them. I know that they wished to have the SD14 shipping before Christmas, but they ended up having to make a hardware fix in the camera that delayed the shipping until this month.
I would expect that you would be able to get your hands on a DP1 by summer (end of June), barring any unforseen major circumstances. They had working pre-production units there, but were not letting images out because they are still working on the imaging processing pipeline.
-
I see how the bricks look weird at the edges but I wonder wether it is camera body artifacts.
Of course it is. How could a lens cause a snap-to-grid effect? A lens is an analog device, it only errs in an analog fashion, totally unrelated to pixels.
Just doesnt look like aliasing to me. As far as the jaggies are concerned I can only tell you about my own: Its mostly user error due to not setting sharpening to -0.8 in the raw converter.
[a href=\"index.php?act=findpost&pid=106040\"][{POST_SNAPBACK}][/a]
Softening jaggies after capture is not the same thing as never capturing them in the first place; same for any kind of aliasing. An AA filter (or a lens that effectivcely acts as one) filters away frequencies above the nyquist, so they are never captured. An unfiltered capture captures frequencies above the nyquist, and after capture, they are indistinguishable from other "mirror" frequencies below the nyquist. You have to filter away valid frequencies to remove the aliased ones.
There is a strange idea going around that a digital camera should be able to record a black pixel next to a white one. It is impossible to do so, and have an accurate recording. What all cameras need for better *real* subject sharpness is more pixels; technology and storage issues are holding that back. With enough pixels, color resolution is no longer an issue with Bayer arrays, and aliasing is no longer an issue (no AA filter needed). Bayer is currently ahead in the miniaturization race.
If the DP-1 had even the mildest of AA filters (which is all it really needs), I would seriously consider buying one (mainly for B&W photography). Without an AA filter, I will not purchase one.
-
John, you put that very well indeed.
Graeme
-
Of course it is. How could a lens cause a snap-to-grid effect? A lens is an analog device, it only errs in an analog fashion, totally unrelated to pixels.
Softening jaggies after capture is not the same thing as never capturing them in the first place; same for any kind of aliasing. An AA filter (or a lens that effectivcely acts as one) filters away frequencies above the nyquist, so they are never captured. An unfiltered capture captures frequencies above the nyquist, and after capture, they are indistinguishable from other "mirror" frequencies below the nyquist. You have to filter away valid frequencies to remove the aliased ones.
There is a strange idea going around that a digital camera should be able to record a black pixel next to a white one. It is impossible to do so, and have an accurate recording. What all cameras need for better *real* subject sharpness is more pixels; technology and storage issues are holding that back. With enough pixels, color resolution is no longer an issue with Bayer arrays, and aliasing is no longer an issue (no AA filter needed). Bayer is currently ahead in the miniaturization race.
If the DP-1 had even the mildest of AA filters (which is all it really needs), I would seriously consider buying one (mainly for B&W photography). Without an AA filter, I will not purchase one.
[a href=\"index.php?act=findpost&pid=106591\"][{POST_SNAPBACK}][/a]
While lumenescence aliasing can be distracting in some cases with the x3 sensor (and no AA filter), it usually provides better, more pleasing output than a CFA sensor with an AA filter---the combination leads to many more issues, especially in the area of color detail. In casual conversations with Foveon, they have tried AA filters over their sensor and the verdict apparently was the images were better without.
CFA technology is definitely more mature, and they are not only ahead of the miniaturization race, they are at their limit (with regards to image performance). That's the nice thing about the "less mature" x3 technology---they still have some headroom to grow. Eventually they may get to the 10 x3 (30 MP) crop sensor, and a 20 x3 mp (60 MP) FF sensor. It's gonna take some hefty pipeline processing to get those to run as fast as a Mk III, however.
-
Of course it is. How could a lens cause a snap-to-grid effect? A lens is an analog device, it only errs in an analog fashion, totally unrelated to pixels.
That is clear to me. The question is not body or lenses here but body or computer. The Artifacts can very easily occur when processing the raw, particularly with sharpening methods that work well with bayer cams.
Then, as I said before, the sharpening slider in the Raw software sitting at zero does not mean no sharpening is applied. As some folks found out, it needs to be set to -0.8 Otherwise you start out in PS with artifacts caused by wrong treatment of the raw file. CAs and stuff like that would in fact be sharpened prior to any removement attempts or resizing.
Softening jaggies after capture is not the same thing as never capturing them in the first place; same for any kind of aliasing. An AA filter (or a lens that effectivcely acts as one) filters away frequencies above the nyquist, so they are never captured. An unfiltered capture captures frequencies above the nyquist, and after capture, they are indistinguishable from other "mirror" frequencies below the nyquist. You have to filter away valid frequencies to remove the aliased ones.
Remains the question what is better: Nothing there above nyquist as with bayer cams or at least something, may it be irregular, captured. But again I am working with a DSLR since only 20 months roughly so the answer to this is beyond what I know (and probably dependent on the actual image and viewer).
Um, and another question: What about the artifacts occurring with the bayer capture? WHat about the moire and nasty stuff like that? Its not that a bayer cam is artifact-free, right?
I'd say let us wait for some reviews how both cameras perform in the real world. Taking pictures is more of an art and less of a science IMO. Note how film grain (what one could call the first artifact ever!) can look nasty and inappropiate in some images and totally great in others.
In fact, maybe they will release an add-on AA filter as sort of a bridge solution...
O.
-
If you have the proper AA filter on a Bayer CFA sensor, you don't get moire nasties. It's funny that on the Foveon pages, they always show bayer images with no AA filter, which will make for colour nasties on edges.
Also, aliassing and moire are both very hard to remove after shooting, but quite easy to avoid if you have the correct level of optical AA when shooting.
About the worst you get on a Bayer image with optical AA is a bit of softness, and given the vastly greater number of pixels Bayer cameras have, this is generally a non-issue.
And, of course, the colour is so much more accurate due to the use of coloured dyes, rather than silicon depth as a colour filters.
Graeme
-
OK it's not ideal you can argue about the lens speed and AA filter, but at the end of the day there is no other pocket size camera available that promises the results of this Sigma. Every other one is a tiny sensor and often jpg only. I'm looking forward to seeing the camera and results, on paper I wont one, I think they will sell truck loads. I saw prints from the new SLR and they looked very nice.
Kevin.
-
If you have the proper AA filter on a Bayer CFA sensor, you don't get moire nasties. It's funny that on the Foveon pages, they always show bayer images with no AA filter, which will make for colour nasties on edges.
Also, aliassing and moire are both very hard to remove after shooting, but quite easy to avoid if you have the correct level of optical AA when shooting.
About the worst you get on a Bayer image with optical AA is a bit of softness, and given the vastly greater number of pixels Bayer cameras have, this is generally a non-issue.
And, of course, the colour is so much more accurate due to the use of coloured dyes, rather than silicon depth as a colour filters.
Graeme
[a href=\"index.php?act=findpost&pid=106859\"][{POST_SNAPBACK}][/a]
Actually, you lose quite a bit of color resolution (color detail) with the "correct" (read high strength) AA filter. This is easily born out looking at test shots of finely detailed color samples under controlled studio conditions. Some camera manufacturers have either eliminated the AA filter (Kodak, Leica), made it optional (Kodak's older DCS series), or made it weaker (D70, D80). This gives you back SOME of the color resolution, but at the expense of color moire. If you remove moire with in software, you'll screw up color resolution in the rest of your image (unless you work very selectively---which is laborious).
With the Foveon chip, you rarely (if ever) see color moire. You will definitely see luminence aliasing (a B&W houndstooth pattern not being accurately reproduced), but it in general will not be as objectionable as the color moire you would get with a CFA sensor (or the mush you would get with a highly AA'ed CFA sensor).
With many subjects, the effects aren't as pronounced---but many of the Foveon-koolaid drinking photographers (myself included) like the "look" of a Foveon image over many CFA images. Another major advantage with the Sigma images is that those "sharp" edges make post production a breeze---including very large prints (you would have seen several if you attended PMA or Focus).
The comment you made about dyes vs depth of light penetration in silicon is just wrong. Both are very reasonable ways to determine color, no method is inherently more accurate. The current advantage of Foveon's approach is that the color sensors are linearly colocated---so a "pixel" (in x,y terms) has 3 color information, as opposed to just 1 color in a CFA imager. 3 chip cameras that use beam splitters (Foveon did that in their early days) also work very well.
-
OK it's not ideal you can argue about the lens speed and AA filter, but at the end of the day there is no other pocket size camera available that promises the results of this Sigma. Every other one is a tiny sensor and often jpg only. I'm looking forward to seeing the camera and results, on paper I wont one, I think they will sell truck loads. I saw prints from the new SLR and they looked very nice.
Kevin.
[a href=\"index.php?act=findpost&pid=106936\"][{POST_SNAPBACK}][/a]
Kevin,
The prints from the SD14 are indeed, very nice. And I believe that Sigma has very reasonable expectations for this camera---if they sell "truckloads" (say the 200K that was mentioned before), they would be blown away.
-
OK its one thing to discuss the different technologies and argue about to AA or not to AA - I have too little knowledge about those things to engage into a serious argument.
ANother thing is looking at actual photos. Maybe you will find these here interesting.
I made those pictures yesterday with my old an trusted SD10, over which the SD14 and the DP1 are said to have been improved a lot sensor-wise.
Both are original-size output from the raw converter, unsharpened, uncurved, unsomethinged.
Only thing I did in PS was reduce a bit of lens-related fringing on the smaller image (its a 100% crop), change color space to sRGB on both images and thats that.
Feel free to download and peek at it. As I said its completely unsharpened, depending on your taste you might want to add a tad of sharpening (maybe 20% and 1pixel Radius in Smart Sharpening) or change the colors a bit, did nothing on those.
Maybe someone can show me those nasty artifacts also.
Images here:
http://www.flickr.com/photo_zoom.gne?id=423098427&size=o (http://www.flickr.com/photo_zoom.gne?id=423098427&size=o)
http://www.flickr.com/photo_zoom.gne?id=423097304&size=o (http://www.flickr.com/photo_zoom.gne?id=423097304&size=o)
REgards,
O.
-
Well, if you look at a raw foveon image in it's raw state, it has very little colour and needs some very strong calculations in the colour matrix to make it colourful, whereas bayer cfa images only need a slight matrix, usually adding just a bit of saturation to get a very colourful image. I think that shows that using silicon as a colour filter is not the best of approaches if colour accuracy is your goal.
Also, bayer cfa cameras generally have much much more resolution than foveon is capable of, so even if you have to downsample, you're still ending up with a larger image.
Interesting to talk of mush - that's just how the human eye sees things - they get blurry and indistinct as they go into the distance. We don't see false aliassing detail as things get small. We see straight lines as straight lines, not stair-steps.
Graeme
-
Well, if you look at a raw foveon image in it's raw state, it has very little colour and needs some very strong calculations in the colour matrix to make it colourful, whereas bayer cfa images only need a slight matrix, usually adding just a bit of saturation to get a very colourful image. I think that shows that using silicon as a colour filter is not the best of approaches if colour accuracy is your goal.
Graeme
[a href=\"index.php?act=findpost&pid=107021\"][{POST_SNAPBACK}][/a]
Graeme,
If you look at either "set" of data in their raw state, neither have color. The readings are voltages proprotional to the number of photons absorbed. The CFA model gets the color from the fact that a pixel is covered by a "green" filter, so it must be 210 worth of "green". It then borrows values from it's red, blue, and green neighbors to get the color.
In the foveon model, the same thing happens for a "pixel", except the photosites that are used to interpolate the color are all spatially co-located. That's a good thing.
If color accuracy is your goal, you should shoot an appropriate target under lighting conditions that you will use, process the RAW files in a way that you deem appropriate, and then measure the results. The Sigma cameras do pretty well.
Another piece of misinformation is:
Also, bayer cfa cameras generally have much much more resolution than foveon is capable of, so even if you have to downsample, you're still ending up with a larger image.
First, a sensor doesn't have "resolution". It has pixels. It's an important distinction. In addition, while it's believed that Bayer sensors are near their limit of usefulness pixel density (let's say 16-20 MP for a full frame sensor), Foveon sensors aren't. If you have a few million dollars (I'll toss in a few as well), I'm sure that Foveon would be happy to crank out a FF sensor based on the current pixel pitch would give you a 14.2 "MP" sensor (which would be called a 43 MP sensor). There's nothing inherient in the technology that limits them from doing that. It's a lot of data to push around, but they had a 48 MP (16 x3) prototype chip in 2000.
Currently, you will end up with much more x,y pixels if you shoot a 5D than if you shoot a SD14. The pixels in the 5D are not as high quality however---there are tradeoffs. B&W resolution will definitely be better with the 5D. Acutence will always be better with the SD14. Color resolution will be better under many situations with the SD14.
Interesting to talk of mush - that's just how the human eye sees things - they get blurry and indistinct as they go into the distance. We don't see false aliassing detail as things get small. We see straight lines as straight lines, not stair-steps.
Indeed, it's really interesting to talk about visual perception and how the eye works. The eye has amazing dynamic range, the ability to pick out fine detail, and a great processing system behind it. Even with a crappy lens! (As an aside, I had my eyes checked and dialated about a month ago---talk about CA with a wide open lens!!!)
One of the things that Foveon does very well is to hold fine detail, even if it's "false". Fine hair in portraits. Fishing line from a fishing pole. Fine fabric detail. The stairstep issues that are often brought up are normally not issues in printing---you don't see them under 99% of the conditions. Where they are issues, one can selectively blur in your favorite tool.
Compare to CFA images---they're non-selectively blurred by the AA filter, and then again by color interpolation. That's one reason for Bruce Fraser's immensely popular sharpening treatise---to regain some of that lost acutence. The reason I love the SD10 (and the SD14 that I've played with) is because making big prints is so easy compared to CFA cameras. The workflow is much easier. And I personally think the prints are better.
I'm not attempting to talk you out of your favorite brand of camera---but there are inherent differences in the Foveon technology that make it superior to CFA systems as a capture technology. This would be equally (or moreso) true for any full color systems. Scanning backs are terrific. 3-chip cameras are too. And perhaps Canon, Nikon, Dalsa, Kodak, or Sony will come out with something full color in the future---but right now, the x3 chip is the only game in town.
-
The problem with this specialized a product is that everyone (almost) will have at least one show stopper. For me, it's not the f4, but rather the lack of articulated screen. I'd use this for unobtrusive street shooting, but after the Sony R1 I'll never give up an articulated LCD. Just like I won't settle for no raw or more noise.
[a href=\"index.php?act=findpost&pid=105681\"][{POST_SNAPBACK}][/a]
Hey Tim,
I remember when we met last year, you were shooting with the R-1. Do you still have it? To me, that camera has so much to like, beginning with the lens and sensor but I am getting a bit camera poor at the moment. I still have the D60 and have added a 30D and more recently the little G7. But there is something very appealing about the R-1. I guess I'm really holdiing out for a lottery win, and the new and improved Leica M-8a or b. But for now, I think I'll just try to find more time to shoot with what I have.
Real reason for the post - Have you seen Michael's new gallery yet? How hard is it going to be for we poor old West Virginians to find when it opens. Betty and I are looking forward to our next trip to Toronto and Michael's next show.
Bill Evans (AKA: Bill in WV)
-
If you look at either "set" of data in their raw state, neither have color. [a href=\"index.php?act=findpost&pid=107042\"][{POST_SNAPBACK}][/a]
I don't think you understand what he was saying; *when* you convert the RAW data to color, literally, without any saturation changes or hue shifts, the Bayer CFA shows much more distinction between colors than the Foveon RAW data does. All three Foveon layers respond almost identically to colors in the blue to green range. Colors in that range record as grey even in the colored version of the RAW data. Only reds, oranges, and yellows have much saturation in Foveon RAW data. Blues and greens, especially the in-between range, are extrapolated and greatly magnify the system and photon noise, leading to the chromatic noise that you see in sky and sea. Deep shadows often have huge off-color blotches when you push them in color, due to doping inconsistencies. Foveon recommends a filter to be used with their sensor, which would make the overall system more discriminating with color, but Sigma does not use it, as it would reduce the sensitivity of the camera greatly.
-
I feel that the automotive industry provides some good insight about the future of bayer vs foveon like devices.
There are better formulas than our rotating piston engines, but car manufacturers and suppliers have invested so much into the our current engines that they just are not interested in taking risk with something else since the benefits might not be huge for customers.
I personnally don't see Canon, Sony, nikon and the other guys jump shipping to a new sensor technology any time soon. Besides the original 1ds, Canon has been extremely conservative in all its DSLR releases eversince, Nikon isn't more of a front runner.
Sigma has adopted a interesting approach and I really hope that the risk they have taken will be rewarded somehow.
If I were Foveon though, I would focus on MFDB sensors...
- Those backs still have large photosites,
- Their users are used to non AA filter side effects,
- High ISO isn't too important there,
- At least some of the users are experienced enough to see the value of having true color capture instead of Bayer,
I guess that there might be some technological issue with producing larger sensor with the Foveon technology, but that is IMHO where they should concentrate their effort.
Cheers,
Bernard
-
JLK, I look at raw images (in the raw) all day. I spend a lot of time with demosaic algorithms, and know their strengths and weaknesses. Aliassing (visible stair-stepping) is the big weakness of the Sigma cameras. The slight softness on bayer cfa cameras is certainly there, but like any errors, an error of omission (some resolution) is generally less offensive than an error of comission (false detail).
When I talk about the colour of a raw image, you know exactly what I mean, so no need to say a raw image doesn't have colour. We all know that the colour matrix coeficients in a foveon camera are going to have to be a lot greater than a bayer cfa camera to generate accurate colour, and any gain you do in a matrix is going to increase noise and other nasties.
I'd hardly say bayer CFA sensors have reached their limits. I would say that silicon being a poor colour filter is a limit that has to be worked around for foveon to improve. Also, imagine how much better a foveon sensor would be if (like in 3ccd video cameras) they offset the red and blue array from the green and use that spatial information to increase resolution.
Graeme
-
Again, since nobody seemed to have noticed.
Can somebody show me the artifacts in those two images made with the older SD10? I am having a hard time here finding some. And I would be interested in an assessment about in which regard a bayer camera would capture color nuances better.
On a side note, I severely struggle to find the chroma noise in the shadows as well. Must be my monitor. Note this is unphotoshopped unsharpened raw converter output of an old Sigma DSLR.
Images here:
http://www.flickr.com/photo_zoom.gne?id=423098427&size=o (http://www.flickr.com/photo_zoom.gne?id=423098427&size=o)
http://www.flickr.com/photo_zoom.gne?id=423097304&size=o (http://www.flickr.com/photo_zoom.gne?id=423097304&size=o)
Regards,
O.
-
Oliver,
No one can view those images; they are set to Private. Besides it's irrelevant since those are just 2 images which is hardly enough to be telling of a cameras performance. And, not all images will display artifacts. The third issue is that in order to compare which two cameras can "capture color nuances better" you would have to capture the same scene in the same light with both cameras.
-
Oliver,
No one can view those images; they are set to Private.
OOps I am sorry. Wasnt aware that this affects direct linking as well. I changed it, again, here are the links.
http://www.flickr.com/photo_zoom.gne?id=423098427&size=o (http://www.flickr.com/photo_zoom.gne?id=423098427&size=o)
http://www.flickr.com/photo_zoom.gne?id=423097304&size=o (http://www.flickr.com/photo_zoom.gne?id=423097304&size=o)
Besides it's irrelevant since those are just 2 images which is hardly enough to be telling of a cameras performance. And, not all images will display artifacts.
There are tons of more images on Pbase, FLickr and on Sigma's website everybody is free to look at. In this thread, an impression arose of Foveon images as full of artifacts and with less detail and worse color than a usual bayer image. I thought it might be interesting for some readers who have not looked at Sigma images so far to get an idea about how the Photos look straight out of the Raw converter without any processing. Maybe some will see certain aspects (capture of color nuances etc) they are missing in other cameras.
The third issue is that in order to compare which two cameras can "capture color nuances better" you would have to capture the same scene in the same light with both cameras.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=107192\")
Even in such a "test" somebody would find a flaw..different sensor sizes..raw converters..lenses..etc. Therefore, to test anything wasnt exactly my intention but rather to provide an illustration after all this talk about nasty artifacts, lack of resolution and wrong colors. People can download these images, look at them, print them or whatever and draw their own conclusions. I see nothing wrong with that.
Sorry if it is of no interest to you. Maybe you will enjoy this here, though:
[a href=\"http://www.ddisoftware.com/sd14-5d/]http://www.ddisoftware.com/sd14-5d/[/url]
O.
-
I definately see a fair bit of stair-stepping in the second of the two photos you link to.
Graeme
-
I definately see a fair bit of stair-stepping in the second of the two photos you link to.
Graeme
[a href=\"index.php?act=findpost&pid=107202\"][{POST_SNAPBACK}][/a]
This camera has to be compared with the other point and shoots on the market, so colour this or that or artifact x or y needs to be seen in context with it's opposition, which is not DSLR's. I don't mean price competitors either, I mean a small camera that goes with you all the time. I don't think for one minute it will do everything or be perfect at what it does do. I can see it being better than the rest on offer though. i hope it works well and Sigma add to the range with other lens choices, I think this could out sell the the sd14 for Sigma. In fact I think when others see the demand for a quality P&S, Canon, Nikon etc will be rushing something to the market.
Kevin.
-
Indeed, think of a fixed 35 mm VR f2.8 lens, weather sealed compact body on top of a Nikon D80 or Canon 400D sensor... with RAW capture.
Sweet thought! I'd buy one instantly for 600 US$.
Cheers,
Bernard
-
... i hope it works well and Sigma add to the range with other lens choices, I think this could out sell the the sd14 for Sigma...
Kevin.
[a href=\"index.php?act=findpost&pid=107470\"][{POST_SNAPBACK}][/a]
Indeed, I'm sure that is Sigma's expectations. And if it does well enough to bring additional SLR shooters into the SD14 (or next gen) dSLR fold, it will have an additional benefit.
My feeling is that this is Sigma's "budget" dSLR answer to the dRebel and D40.
-
Indeed, think of a fixed 35 mm VR f2.8 lens, weather sealed compact body on top of a Nikon D80 or Canon 400D sensor... with RAW capture.
Sweet thought! I'd buy one instantly for 600 US$.
Cheers,
Bernard
[a href=\"index.php?act=findpost&pid=107576\"][{POST_SNAPBACK}][/a]
Oh yes please, I can't see why they have not done this already. I wouldn't mind a wide camera, say Canon 16mp and a fixed super quality 20mm or wider with built in rise would be nice too. Hassel had the Super wide, others have made a quality wide so we know there's a market. I would pay lots more than $600.0, x by 10 and there would still be many buyers.
I hope this Sigma delivers.
Kevin.
-
Or you could slip in a small sliver of glass, an anti-alias filter which really is an engineering necessity, take the resolution hit, but get smooth, continous images.
Graeme
[a href=\"index.php?act=findpost&pid=106044\"][{POST_SNAPBACK}][/a]
Well if you're willing to take the resolution hit of an AA filter, just use Gaussian blur in Photoshop. Spreading the photons or blurring the RGB levels are pretty equivalent with a Foveon sensor.
And leave the sharp, slightly aliased pictures to those who don't mind it. In practice as long as you don't have strong contrast, the aliasing seems not detectable.
-
Softening jaggies after capture is not the same thing as never capturing them in the first place; same for any kind of aliasing. An AA filter (or a lens that effectivcely acts as one) filters away frequencies above the nyquist, so they are never captured. An unfiltered capture captures frequencies above the nyquist, and after capture, they are indistinguishable from other "mirror" frequencies below the nyquist. You have to filter away valid frequencies to remove the aliased ones.
You can find a sensor able to record frequency above the Nyquist frequency?? Maybe you should reread your Signal Processing textbook, no? What happens is that the Foveon is able to record signal at Nyquist frequency whereas a standard Bayer cannot, due to the low-pass AA filter in front of the sensor. Whether it is desirable or not to resolve Nyquist frequencies is another question. The jaggies would probably not annoy you if the pixel resolution was 3x higher as in a classic Bayer. After all, the only thing that counts is whether you can see them in print or at usual display magnification.
There is a strange idea going around that a digital camera should be able to record a black pixel next to a white one. It is impossible to do so, and have an accurate recording.
I'm curious to know why...
What all cameras need for better *real* subject sharpness is more pixels; technology and storage issues are holding that back. With enough pixels, color resolution is no longer an issue with Bayer arrays, and aliasing is no longer an issue (no AA filter needed). Bayer is currently ahead in the miniaturization race.
With a Bayer array, AA filtering is always necessary to distribute light over neighboring photosites, otherwise you're not able to differentiate color close to Nyquist frequency. In other words, you are not able to tell the difference between a tiny white reflection and a colored one. No AA filter = color artifacts.
-
A post blur of a sampled image is NOT the same as a pre-blur before sampling. That's he problem with aliases - once they get into the system they cannot be got out easily. If they could be removed by a small gaussian blur, we'd all not be bothered by aliases, and we'd just post-blur when necessary. But it doesn't work like that. I'd remember that Foveon themselves say that their sensor still needs an OLPF but Sigma ignore this.
Graeme
Well if you're willing to take the resolution hit of an AA filter, just use Gaussian blur in Photoshop. Spreading the photons or blurring the RGB levels are pretty equivalent with a Foveon sensor.
And leave the sharp, slightly aliased pictures to those who don't mind it. In practice as long as you don't have strong contrast, the aliasing seems not detectable.
[a href=\"index.php?act=findpost&pid=174490\"][{POST_SNAPBACK}][/a]
-
Again, since nobody seemed to have noticed.
Can somebody show me the artifacts in those two images made with the older SD10? I am having a hard time here finding some. And I would be interested in an assessment about in which regard a bayer camera would capture color nuances better.
On a side note, I severely struggle to find the chroma noise in the shadows as well. Must be my monitor. Note this is unphotoshopped unsharpened raw converter output of an old Sigma DSLR.
[a href=\"index.php?act=findpost&pid=107134\"][{POST_SNAPBACK}][/a]
You've used a very strong 'effective' AA filter in these shots; out-of-focusness. Many of the shadows are OoF and there are only small areas in each shot that are in focus.
Nevertheless, I don't see any major problem with artifacts in any of the Foveon images that I've viewed on my monitor, and as we all know, users of MFDBs seem to be very pleased with the superior results they get, claiming that a 20mp DB without AA filter produces noticeably superior results to a 21mp 35mm dslr.
-
A post blur of a sampled image is NOT the same as a pre-blur before sampling.
Graeme
[a href=\"index.php?act=findpost&pid=174516\"][{POST_SNAPBACK}][/a]
Sorry Grame I don't see your point. Imagine photons falling into bins. You can distribute them with a physical filter over a larger area. Then your photon counts will be exactly those you'd have without filter, convoluted by the mathematical filter (kernel) that matches your physical LPF (it may not be exactly Gaussian of course). If your physical LPF is not dispersive (which is to be hoped for) then it's the same filter for all RGB levels.
I don't see how applying a convolution on RAW data can be any different from having a physical LPF, (if we neglect the slight difference due to photons falling on photosite borders, of course).
I bet the differences you are reporting are due to blurring PROCESSED, sharpened images.
Yes there is natively a small artificial contrast induced by photosite borders, but if your raw converter allows applying a slight preblur before any other treatment, then you shouldn't see any aliasing at all. That's something fairly easy to implement in a raw converter, especially if users ask for it. So I don't see why having no physical LPF is a design flaw.
If you can show me evidence of the contrary, I'm ready to change my mind. :-)
Anyway applying low sharpening for Foveon images should solve your problem in practice. The only reason Sigma oversharpen them is to visually match 12 MP images. I don't mind slighty lower subjective resolution, especially for a compact camera. And the lower pixel count means I can put twice as many images on a memory card!
-
It's very , very different. Blur then sample is very different to sample then blur! That's because aliasing happens at the point of sampling when too high frequencies enter the system. If you stop them arriving at the sampling, aliasing and moire never occur. If you let them through sampling, they fold back into the image an appear at lower frequencies (hence the name alias) so that the amount of blur needed to remove them, especially in the case of moire, will remove all detail from the image. I'd read through Alan's post here: http://forums.dvdoctor.net/showthread.php?...hlight=aliasing (http://forums.dvdoctor.net/showthread.php?t=42116&highlight=aliasing) about aliasing as that will explain the situation much better than I can.
But put it this way - an OLPF is expensive, and software is cheap. If you thought that software blurring could do the same as a physical OLPF, don't you think camera manufacturers would do so? It would be much cheaper, and totally user configurable from full through none, and would make everyone really happy! But, you see, it doesn't work that way, so we have to do it properly and expensively with a real physical before the sensor filter.
Graeme
-
Here is a simple example of the difference between "blur then sample" and "sample then blur". Imagine a thin line of illumination running diagonally across a sensor. To be definite, lets say the line starts at the corner of a pixel and traverses two pixels horizontally for one pixel vertically. An optical system with no blur will produce an image with a "staircase" pattern. Now blur the line by a fraction of a pixel. This will spread some light into pixels barely touching the original line, partially filling in the corners of the staircase pattern. But note that there will be no light added to pixels which do not contain the blurred line. Contrast this to the case where the image is blurred after sampling. The corners of the staircase pattern will again be filled in, but light will also be added to more distant pixels which do not contain the blurred line.
-
Here is a simple example of the difference between "blur then sample" and "sample then blur". Imagine a thin line of illumination running diagonally across a sensor. To be definite, lets say the line starts at the corner of a pixel and traverses two pixels horizontally for one pixel vertically. An optical system with no blur will produce an image with a "staircase" pattern. Now blur the line by a fraction of a pixel. This will spread some light into pixels barely touching the original line, partially filling in the corners of the staircase pattern. But note that there will be no light added to pixels which do not contain the blurred line. Contrast this to the case where the image is blurred after sampling. The corners of the staircase pattern will again be filled in, but light will also be added to more distant pixels which do not contain the blurred line. (I tried to attach images illustrating this, not sure if they will make it.)
-
(I tried to attach images illustrating this, not sure if they will make it.)
[a href=\"index.php?act=findpost&pid=174620\"][{POST_SNAPBACK}][/a]
I saw the attachment earlier today, but it is gone now.
-
You have a point, but a weak one. You're telling me that frequencies *above* Nyquist frequency will translate to artifacts (folding back to lower freqs) if you haven't filtered them out. That's true. These artifacts will be luminance artifacts (and not moire, since the Foveon will never give you color artifacts). Your example with the fine line is exactly similar, you're speaking frequencies above Nyquist. But you should think about the size of your 'fine line'.
A Foveon sensor has photosites < 8 micrometers wide, which on a 20.3 x 13.7 mm sensor translates into a Nyquist freq of 64 lp/mm. Now if you're shooting with super-heavy tripod and leica glass, you may encounter this limit. Normal people, hand-held camera, sigma lens... it'll be far enough low-pass filtering, you'll never reach strong signal at 40 lp/mm. So you may think it's stupid not to put a OLPF in a DP1, but for me it's just common sense.
The only real problem with the DP1 how they'll handle the color noise. The rest is phantasm.
-
The only real problem with the DP1 how they'll handle the color noise. The rest is phantasm.
[a href=\"index.php?act=findpost&pid=175500\"][{POST_SNAPBACK}][/a]
The images *ARE* aliased, despite all of your theorizing to the contrary. How can you miss it? Facial hair stubble that appears in some cases as a slight darkening of two lines of pixels, and in other cases, solid pixels of hair, exactly one pixel wide, and exactly two pixels long, and in the presence of other such stubble, in an obvious grid with no sub-pixel precision. Happens quite frequently. Aliasing is rampant in Sigma images.
-
The images *ARE* aliased, despite all of your theorizing to the contrary. How can you miss it? Facial hair stubble that appears in some cases as a slight darkening of two lines of pixels, and in other cases, solid pixels of hair, exactly one pixel wide, and exactly two pixels long, and in the presence of other such stubble, in an obvious grid with no sub-pixel precision. Happens quite frequently. Aliasing is rampant in Sigma images.
[a href=\"index.php?act=findpost&pid=175535\"][{POST_SNAPBACK}][/a]
I'm simply saying:
1. The absence of anti-alisaing filter will never be a problem for the DP1
2. The aliasing you complain of is mainly due to *oversharpening* images in order to make them appear as sharp as 12 MP images.
As I suppose you routinely print 20 x 13.3" prints that you look at 10 inches, you're certainly very worried about this. I am not, and I reckon many people here aren't either.
But before I can take your comments seriously, you'll have to shoot with the SD14 (or DP1, if ever it comes), set the image processing parameters to what suits you the most, and be still dissatisfied with the results.
I suspect you have never hold an SD14 in your hands, never tried to, and never printed a Sigma image to any size. Am I wrong?
-
I'm simply saying:
1. The absence of anti-alisaing filter will never be a problem for the DP1
It already is for the SD14, and the DP1. They have too few pixels to have aliased pixels.
2. The aliasing you complain of is mainly due to *oversharpening* images in order to make them appear as sharp as 12 MP images.
No. Aliasing due to over-sharpening is different from aliasing due to aliased capture.
As I suppose you routinely print 20 x 13.3" prints that you look at 10 inches, you're certainly very worried about this. I am not, and I reckon many people here aren't either.
I can see the aliasing with the images downsampled to fill my monitor. That's all I need to have a problem.
But before I can take your comments seriously, you'll have to shoot with the SD14 (or DP1, if ever it comes), set the image processing parameters to what suits you the most, and be still dissatisfied with the results.
You don't seriously think the JPEG settings are going to affect the aliasing of the capture, do you?
I suspect you have never hold an SD14 in your hands, never tried to, and never printed a Sigma image to any size. Am I wrong?
[a href=\"index.php?act=findpost&pid=175545\"][{POST_SNAPBACK}][/a]
You don't have to handle a camera to tell things about its IQ. Are you serious? You're clutching at straws.
I'd be very embarrassed if that was what I had to resort to in order to defend my position.
But I have printed Sigma samples, and I can see the aliasing in them.
It seems that most people are blind to aliasing. It's not just Jaggies - that's just one artifact of aliasing. There is a very general, and obvious (to me) redistribution of the analog image in a snap-to-grid effect that makes aliased captures look unnatural, and incompletely sampled.
-
I can see the aliasing with the images downsampled to fill my monitor. That's all I need to have a problem.
In these conditions there's no discussion possible. The artifacts are in your head.
-
As far as I've seen, the artifacts are in just about every in-focus Sigma image I've seen when viewed 1:1.
Graeme
-
As far as I've seen, the artifacts are in just about every in-focus Sigma image I've seen when viewed 1:1.
[a href=\"index.php?act=findpost&pid=175712\"][{POST_SNAPBACK}][/a]
And 1:1 with 4.6M pixels is not that far away from full-screen scaling with hires monitors.
-
In these conditions there's no discussion possible. The artifacts are in your head.
[a href=\"index.php?act=findpost&pid=175683\"][{POST_SNAPBACK}][/a]
I'll continue anyway, without you. The main problem I have with aliased capture is the lack of sub-pixel precision, or the "snap-to-grid" effect. This does not improve tremendously with minor ratios of downsampling, because even though the downsampling increases sub-resultant-pixel precision, the degrees of freedom are still restrained. I estimate about 25%-30% linear resizing is necessary so that the snap-to-grid effects are lost. My monitor has a height of 1050 pixels. The Sigma images are 1760 pixels high. That's a downsampling of 1050/1760, or 60%.
Back to one of the points you made earlier; the issue of blurring before or after capture:
1) You are overlooking the very pertinent fact that cameras without AA filters have blind space between the pixels; cameras with AA filters do not. This, in itself is one of the major contributors to aliasing. A true box capture with no spatial waste aliases far less than even one that captures 90% of the area effectively. Two diagonal-neighbor pixels can easily miss a thin line passing through them completely. Even if the area that is totally blind is very small, the fringes can be less sensitive and still have more weighting towards the middle, so even with microlenses, vestiges of point-like sampling can still be present.
2) Blurring after capture can not reduce low-frequency aliasing effects like picket fences that change in luminance over their run.
I am currently working out a method of doing simulations of the issues we are discussing in PS, including the dead space issue.
-
As far as I've seen, the artifacts are in just about every in-focus Sigma image I've seen when viewed 1:1.
Graeme
[a href=\"index.php?act=findpost&pid=175712\"][{POST_SNAPBACK}][/a]
Right. What I'm saying is that precisely I don't view image crops at 100%. I view them in prints. At normal distance. Without a magnifying glass.
I don't expect my setup to give me 60 lp/mm. I don't own a 10k$ Leica. This thread is about a $700 compact camera (for those who forgot).
Besides, some people here should understand that setting parameters in the image processing pipeline goes beyond chossing a jpeg compression level (personally I never shoot jpeg). From the raw cmos signal to the image, there are a bunch of transformations, and in all cameras I know you can set your personal sharpening level. Those who complain about jaggies should first check whether this is the solution to their problem (if it is real). In any case if you've to print a giant poster, a 0.5 Gaussian blur in Photoshop will be quite efficient without affecting much the overall sharpness.
When the Kodak DCS 14n appeared without anti-aliasing filter, pixel experts were duly horrified (" 't was a bold but stupid move"). When the Nikon D70 was released with an OLPF that was in the high cutoff range, example of situations where moire appears flourished on the web (and this is far worse than jaggies, because the artifactual color is really apparent at the scale of > 10 pixels. To own one, I know this is *really* a marginal problem (the small and moderately bright viewfinder is far more annoying, but that's not visible on a website image viewed at 1:1...). When the Leica M8 appeared, well I dunno what our expert pixel peepers thought (probably "heck, no OLPF, but I can't really say on a forum that Leica engineers are clowns, so let's forget about it").
Now it's the turn of the Foveon X3 and its derivatives. Come on guys, please talk about color rendition, color profiling, ability to render skin tones, highlight details, max sensitivity, color and luminance noise, whatever, but let the 100% crop fantasms to those that never take pictures (or only photograph resolution charts), and consequently never show them.
-
Indeed, think of a fixed 35 mm VR f2.8 lens, weather sealed compact body on top of a Nikon D80 or Canon 400D sensor... with RAW capture.
Sweet thought! I'd buy one instantly for 600 US$.
Ditto. The only other thing I'd look for is a 85mm VR included.
-
I am currently working out a method of doing simulations of the issues we are discussing in PS, including the dead space issue.
[a href=\"index.php?act=findpost&pid=175732\"][{POST_SNAPBACK}][/a]
Here we go:
(http://www.pbase.com/jps_photo/image/93229983/original.jpg)
The basic idea here is that I started with an original with 16x as many pixels, to simulate an analog original, with a very sharp lens with low diffraction and sharp-edged subjects (although all shapes were placed in PS with anti-aliasing enabled). All are immediately-rasterized shapes on a single layer in PS, except the strip on the right of variable frequency, which I generated mathematically in Filter Factory. The appended "AA" means that a gaussian blur equal to 1/2 the resulting pixel width was applied before sampling. "Box" means the sampling of each 4x4 tile of the original included all 16 pixels in the output pixel, "Vpoint" means that a 3x3 tile only was used, and there was consequently some dead space in the sampling.
Hopefully, you will see here what it is that I object to in aliased sampling.
To my eyes, in order of decreasing sampling quality they are:
1) BoxAA -best but purely theoretical, unless a camera were to use binning of extreme pixel density with an AA strength calculated for the resulting pixel. Requires no dead or insensitive space between pixels.
2) VpointAA - similar to what we would get with a greyscale camera with an AA filter.
3) Box - theoretical unfiltered capture, with no dead or low-sensitivity space.
4) Vpoint - Similar to what we are actually getting with Foveon and greyscale cameras with no AA filter.
The main problem with the unfiltered captures is that they have spatial jitter. The objection that this is only visible with pixel peeping is ridiculous, because the jitter is visible with any amount of blurring, downsampling, or stepping back from the monitor unless such are so extreme that you can't recognize these captured shapes at all. Any attempt to blur (or downsample, or step back from) the unfiltered captures after the fact makes them much softer than the filtered ones, without removing the jitter, and makes them totally incomprehensible if done to the extent that the jitter is no longer noticeable.
Now, it might take these worst-case B&W originals for you to see the jitter, but *I* see it in regular images with lower-contrast subjects, and hopefully you now have an idea of what it is I see in aliased captures that I do not like - false detail that arbitrarily says something which has nothing to do with what is really there.
YES - it lets you know that you have successfully focused on the medium.
NO - it is not the real subject.
-
Here we go:[a href=\"index.php?act=findpost&pid=176335\"][{POST_SNAPBACK}][/a]
Take special note of the strip on the lower right; the original increases in line frequency going down; the low-frequency artifacts from nyquist mirroring in some are there forever, with no chance of removal.
-
It's a very nice demonstration, and shows the aliasing artifacts quite nicely, but it would be more convincing if you didn't assume that the lens resolved 250 lp/mm on a Foveon-type (20 mm wide) sensor.
The day such a lens will be manufactured, I'm not sure you and me will be alive (if it ever happens)...
-
The Foveon in the SD14 is 20.7mm wide, and has 2640 pixels, therefore it's resolution is 127p/mm or about 63 lp/mm.
Graeme
-
The Foveon in the SD14 is 20.7mm wide, and has 2640 pixels, therefore it's resolution is 127p/mm or about 63 lp/mm.
[a href=\"index.php?act=findpost&pid=176403\"][{POST_SNAPBACK}][/a]
Shift everything by 1/2 pixel, and it won't resolve it at all, though. 0 contrast.
-
Shift everything by 1/2 pixel, and it won't resolve it at all, though. 0 contrast.
[a href=\"index.php?act=findpost&pid=176450\"][{POST_SNAPBACK}][/a]
In my examinations of luminance resolution tests amongst cameras, I've found that both bayer cameras and Foveons have a consistent, reliable maximum resolution (the only kind there really is, IMO; relying on luck is not "resolution" in my mind), is about the same, relative to the pixel frequency.
-
It's a very nice demonstration, and shows the aliasing artifacts quite nicely, but it would be more convincing if you didn't assume that the lens resolved 250 lp/mm on a Foveon-type (20 mm wide) sensor.
The day such a lens will be manufactured, I'm not sure you and me will be alive (if it ever happens)...
[a href=\"index.php?act=findpost&pid=176399\"][{POST_SNAPBACK}][/a]
... but I see these same artifacts in actual Sigma images; mortar periodically missing, etc, etc.
-
The Foveon in the SD14 is 20.7mm wide, and has 2640 pixels, therefore it's resolution is 127p/mm or about 63 lp/mm.
Graeme
[a href=\"index.php?act=findpost&pid=176403\"][{POST_SNAPBACK}][/a]
And if you assume that the original image is 4x more defined that the sensor grid, as John did, you get 4x63 = 252 lp/mm.
-
And if you assume that the original image is 4x more defined that the sensor grid, as John did, you get 4x63 = 252 lp/mm.
[a href=\"index.php?act=findpost&pid=176631\"][{POST_SNAPBACK}][/a]
There are plenty of lenses that do that in their sweet spots with at least traces of contrast, and as I said, all my original shapes were placed with anti-aliasing. The only thing that isn't anti-aliased is the very bottom of the variable-frequency strip, as I did not over-sample it in the calculations. It is aliased enough to have some spatial jitter, but no nyquist mirroring is visible (it increases in frequency very slightly up until the end).
Regardless, the bottom line is that these types of artifacts are visible in real Sigma images with real lenses.
-
Whilst I understand the concerns about alaising artifacts and false detail from a viewpoint of total accuracy, it seems we have a phenomenon here which could legitimately appeal to personal taste.
If you are attempting to create a work of art, then you manipulate the image to get the effect that satisfies you, and you hope it also satifies the customer if you are attempting to sell the work.
It would seem clear that John Sheehy would not be a potential customer for such a work produced from a Sigma camera because he is so finely tuned to the effects of aliasing artifacts, which he doesn't like, and which I think he would notice and identify as such where most people wouldn't.
Most people might be deceived into thinking that the Sigma image is just plain sharp and crispy. Perhaps they don't care if some of that detail is false. Perhaps here, with the characteristics of the Foveon sensor, we have a type of photographic impressionism; the sharpness of the photograph caricaturised.
-
Most people might be deceived into thinking that the Sigma image is just plain sharp and crispy. Perhaps they don't care if some of that detail is false. Perhaps here, with the characteristics of the Foveon sensor, we have a type of photographic impressionism; the sharpness of the photograph caricaturised.
[a href=\"index.php?act=findpost&pid=176671\"][{POST_SNAPBACK}][/a]
You could look at it that way, but I don't think that this is what the typical person impressed by the spatial aspects of Sigma images sees; I think that they assume that they are really seeing better resolution than what is really there.
Do you really want impressionism to happen at the capture level? You can do much more with an image aliasing it or mosaicing it yourself, after the fact of capture.
I'm looking towards the future, and it scares me that public demand (which drives marketing to some degree) seems to be for low pixel densities with pseudo-detail. Much more can be done with a higher-resolution sensor. Software aliasing filters would be easy to implement, with all kinds of degrees of freedom; a filter could present a square to you in the middle of its dialogue, and you just draw a luminance mask inside and outside of it to show the weighted influence of the source area to the output pixels. I don't need any camera capturing RAW for me with artificial texture.
-
I'm looking towards the future, and it scares me that public demand (which drives marketing to some degree) seems to be for low pixel densities with pseudo-detail. Much more can be done with a higher-resolution sensor.
Your scare is very weird, because it's not at all backed by the facts.
What scares me is exactly the contrary: artificial demand for more resolution and more megapixels (the more the better, isn't it? - marketing blurb). Once upon a time a company called Fuji made a series a fine compact cameras around a 6.3 MP sensor (F10, F11, F20, F30). These were the first compact cameras that were able to produce decent images at 800 ISO. They sadly were the last too. Fuji dropped this in under the pressure of the pixel-peeping crowds that always demand more pixels. The low-light capable F30 was replaced by a 10 MP F50 which is no better than the rest of the 10-12 MP compact cameras: useless over 400 ISO.
For the Foveon as for the other CMOS/CCD sensors (and as for silver halide emulsions...), there's a trade-off between resolution and sensitivity. It's a good thing Foveon did not choose to increase resolution, because noise would be even more problematic. It is problematic enough to my taste.
I clearly prefer crisp 8x12" prints (or slideshows) with low noise to accurate 1:1 luminance detail on my monitor and useless prints plagued with color noise.
But John, you can rest assured that the vast majority still believes it's normal to have Christmas trees in any forest shot at 800 ISO. And that you need to have enough resolution for 20x30" prints even if the lens resolves 30 lp/mm (24x36mm-equivalent) and less than thousand photographers on earth print a 20x30" each year.
-
Most people might be deceived into thinking that the Sigma image is just plain sharp and crispy. Perhaps they don't care if some of that detail is false. Perhaps here, with the characteristics of the Foveon sensor, we have a type of photographic impressionism; the sharpness of the photograph caricaturised.
Exactly.
As well as most people will be deceived into thinking that the colors of an image derived from a Bayer array sensor are the original ones. In some way the sharpness of a Bayer-type image is also a caricature as it is reconstructed by filtering (unsharp masking etc - which in some situations lead to sharpening artifacts).
All these technologies rely on what kind of approximation of the original light stimulus we are ready to accept as a faithful representation.
-
Because silicon is a poor colour filter, the colours on a Foveon are somewhat guessed too....
Graeme
-
Your scare is very weird, because it's not at all backed by the facts.
What scares me is exactly the contrary: artificial demand for more resolution and more megapixels (the more the better, isn't it? - marketing blurb).
That's not marketing blurb. It's fact, easily demonstrated with cameras that do RAW, while the opposite has no proof at all, just a bunch of mantras, and bad conclusions drawn from the madness of the noise reduction race, and the high RAW-illiteracy rate of people who are otherwise generally on the up-and-up. Back in the day, almost everyone I knew who was serious about photography knew about the darkroom; today, only a very small number of people even look at RAW data and try to work it themselves. Then, they don't know what it is that they need to complain about, blaming the sensor for things done by software.
Once upon a time a company called Fuji made a series a fine compact cameras around a 6.3 MP sensor (F10, F11, F20, F30). These were the first compact cameras that were able to produce decent images at 800 ISO.
Noise reduction, period. No sensor that small has low noise images. The Fuji 6.3MP sensor is average, in its class, perhaps below average. It has a quantum efficiency a stop lower than the much-maligned 10MP Panasonic sensor in the FZ50, and slightly more read noise *per* pixel, with barely more than half of them. Fuji, like Sigma, has created an impressionistic style of NR and conversion that has a strong following.
They sadly were the last too.
They were the last, perhaps, to be rescued in the eyes of the optically naive by noise reduction at a 100% pixel view. When you back away from individual trees, and look at the forest, then you will see why there is value in having assembled all these trees together.
Fuji dropped this in under the pressure of the pixel-peeping crowds that always demand more pixels.
The term "Pixel Peeper", unfortunately, gets two groups of people lumped into it, with very different viewpoints. I get called a pixel peeper because I study noise and signal to some degree, and it involves measurements with pixels, but I do not see the pixel as the determiner of image quality; it is only the quality of a spatial witness with a *varying* degree of accountability, depending on total pixel population in a displayed image, and the usage. The person who complains about the MP race, in general, claiming that it leads to more noise, based on the individual pixel noise and/or softness is the true pixel peeper, IMO, at least as far as the negative connotations are concerned.
The low-light capable F30 was replaced by a 10 MP F50 which is no better than the rest of the 10-12 MP compact cameras: useless over 400 ISO.
Do any of them output RAW? Take the same shot on one and an S6500fd (which has the 6.3MP snesor and RAW), and link to the RAW files, and I will show you both, resampled to the same size, with no NR or other non-homogeneities.
For the Foveon as for the other CMOS/CCD sensors (and as for silver halide emulsions...), there's a trade-off between resolution and sensitivity.
Film is a *completely* different story than digital sensors. Film is less sensitive when the grain is fine, and the developed grain expands very slowly, so it takes a lot of light to get a saturated capture (more original hots; less growth during devlelopment). Digital has low sensitivity by being able to capture lots of photons per unit of area, but slowly.
It's a good thing Foveon did not choose to increase resolution, because noise would be even more problematic. It is problematic enough to my taste.
Most of the noise that you see in Foveons is due to two things:
1) Problems with the color filtering method used leads to the blotches of green vs purple-blue in the shadows, and
2) Shot and read noises are magnified by the mathematical tricks used to get distinct, saturated hues out of the poorly separated RAW color channels.
I clearly prefer crisp 8x12" prints (or slideshows) with low noise to accurate 1:1 luminance detail on my monitor and useless prints plagued with color noise.
What is "accurate 1:1 luminance detail"?
But big mosaics of noise are not a problem? I can not simulate or replicate this alleged benefit you propose with low pixel density; the shot noise has the same energy, it's just a bit shallower, but bigger noise and image "grain". The read noise usually has less energy with a higher pixel density.
But John, you can rest assured that the vast majority still believes it's normal to have Christmas trees in any forest shot at 800 ISO.
I don't make any sense out of that statement. Try paraphrasing it.
And that you need to have enough resolution for 20x30" prints even if the lens resolves 30 lp/mm (24x36mm-equivalent) and less than thousand photographers on earth print a 20x30" each year.
[a href=\"index.php?act=findpost&pid=177783\"][{POST_SNAPBACK}][/a]
Resolution has nothing necessarily to do with print size. High pixel density is also to get rid of artifact in the capture, and generally reduces image-level read noise, while maintaining image-level shot noise. You can resample a 50MP image to 5MP, 4MP, 3MP, 12MP, etc, with a minimum of artifact. A 6MP resampled to the same sizes will have more artifacts, and probably a bit more read noise.
You seem to completely miss the value and beauty of oversampling.
-
Because silicon is a poor colour filter, the colours on a Foveon are somewhat guessed too....
Graeme
[a href=\"index.php?act=findpost&pid=177790\"][{POST_SNAPBACK}][/a]
Yes, that's right. Color balance with anything different than 5000K light seems tricky with the Foveon. A number of CCD dSLR have problems with deep purple hues. And the color spaces of our monitors and photographic papers are also limited... Think also about human perception and remember that three color channels are not always enough to accurately present the colored spectrum to the human eye!
There's no perfect technology in photography (or anywhere else), so all depends on the compromises one's willing (or not willing) to do.
To come back to the original topic, the Sigma DP-1, due next month, this kind of compact camera will be mostly judged on its ability to shoot in low light. That's the main advantage the large sensor is supposed to bring in (the other one --which shouldn't be neglected tough-- is that a small depth of field is possible with the Sigma and not with the standard digicams). The fact that the lens opens at f/4 is clearly a handicap, though there was little to do about this if the main objective was to keep the thing a compact camera (and not a bulky M8-sized one). So the sensor'd better properly handle high ISO (and that, by design, is not the forte of the Foveon). The images shot with the D14 at 800 ISO didn't impress me as far as noise is concerned. With the DP1 I'd like to see an improvement over the D14 at high sensitivity. Whether it'll be there or not is the question.
-
You seem to completely miss the value and beauty of oversampling.
That's perfectly right. Explain me why Fuji was able to produce decent JPEG images at 800 ISO (and still usable ones at 1600) witha 6.3 MP sensor and is now unable to do it with a 10 MP sensor.
Their engineers have Alzeihmer's disease?
They decided that what you claim was an 'impressionistic' style -- and attracted followers, for sure -- was to be ditched because John Sheehy judged it looked bad?
They reasoned that Canon set the standard, and as it is crappy, they'd happily stick to it?
(BTW, as you didn't understand my comment: Xmas trees have nice colored garlands... normal fir trees have not -- at least in the forests I know)
-
They were the last, perhaps, to be rescued in the eyes of the optically naive by noise reduction at a 100% pixel view.
No. [I edited this after realizing the main difference between these cams]
Actually they were rescued by slight overexposure.
Compare this (http://www.imaging-resource.com/PRODS/F30/FULLRES/F30hSLI0800.HTM) (Fuji F30, 800 ISO) with this (http://www.imaging-resource.com/PRODS/SD950IS/FULLRES/SD950IShSLI0800.HTM) (Canon SD950, 800 ISO) and or this (http://www.imaging-resource.com/PRODS/F50FD/FULLRES/F50FDhSL0800.HTM) (Fuji F50fd).
Look at the patterns in the black paper napkin, for example. The F30 gives far better results at 800 ISO. The downside is blown up highlights (white thread). In low light it is very rarely a problem, so there's the trick.
-
With the DP1 I'd like to see an improvement over the D14 at high sensitivity. Whether it'll be there or not is the question.
[a href=\"index.php?act=findpost&pid=177926\"][{POST_SNAPBACK}][/a]
They have exactly the same sensor; I don't think you can expect much difference, except in software NR in the JPEGs.
-
There's something I don't get with the Fuji F30.
At 800 ISO, since the sensor has constant sensitivity, there's absolutely no chance of saturating the CCD wells. So the blown-up highlights come from their tone curve. This means that there's enough signal in the raw data to correctly reconstruct the shades (e.g. the dark napkin), and that in order to have enough contrast they sacrificed the highlights.
If John's theory is right (basically the law of large numbers + uncorrelated read noise, as far as I understand) why then isn't there enough signal in the 10 MP/12MP sensors in the same areas? If you look at the noise in these areas, you guess that even extra-smart processing won't recreate the patterns. It's not just a different tone curve. The F30 has more signal/noise in the shades. So?
-
John's been strangely less eloquent lately...
-
There's something I don't get with the Fuji F30.
At 800 ISO, since the sensor has constant sensitivity, there's absolutely no chance of saturating the CCD wells. So the blown-up highlights come from their tone curve. This means that there's enough signal in the raw data to correctly reconstruct the shades (e.g. the dark napkin), and that in order to have enough contrast they sacrificed the highlights.
If John's theory is right (basically the law of large numbers + uncorrelated read noise, as far as I understand) why then isn't there enough signal in the 10 MP/12MP sensors in the same areas? If you look at the noise in these areas, you guess that even extra-smart processing won't recreate the patterns. It's not just a different tone curve. The F30 has more signal/noise in the shades. So?
[a href=\"index.php?act=findpost&pid=178173\"][{POST_SNAPBACK}][/a]
I think the issue here is that the raw data is the ccd well votage (proportional to charge) multiplied by an amplifier before input to the adc. If the amplified signal exceeds the level required to give all 1s from teh adc then, regardless of the well saturation the highlights are clipped in the output.
Mike
-
There's something I don't get with the Fuji F30.
At 800 ISO, since the sensor has constant sensitivity, there's absolutely no chance of saturating the CCD wells.
It happens all the time with night shots with lights in them.
So the blown-up highlights come from their tone curve. This means that there's enough signal in the raw data to correctly reconstruct the shades (e.g. the dark napkin), and that in order to have enough contrast they sacrificed the highlights.
In the vast majority of cameras, the sensor saturation level is only present in the lowest ISO, if at all. It is the digitization process that normally clips data. Many manufacturers clip the RAW data at their lowest ISO a bit short to keep the non-linear sensor saturation out of range.
If John's theory is right (basically the law of large numbers + uncorrelated read noise, as far as I understand) why then isn't there enough signal in the 10 MP/12MP sensors in the same areas? If you look at the noise in these areas, you guess that even extra-smart processing won't recreate the patterns.
I haven't seen any compact Fuji 12MP RAWs. It is often the case, however, that the higher MP cameras look worse at 100% pixel view, but not at the full image.
It's not just a different tone curve. The F30 has more signal/noise in the shades. So?
[a href=\"index.php?act=findpost&pid=178173\"][{POST_SNAPBACK}][/a]
They don't have any special S/N in the shades; they have aggressive, cartoon-generating noise reduction. There is no texture in fuji compact JPEGs in shadows or at high ISOs.
-
John's been strangely less eloquent lately...
[a href=\"index.php?act=findpost&pid=179252\"][{POST_SNAPBACK}][/a]
I read your post a few days ago, but it was a bit tangential to what I thought we were discussing, and your language really didn't suggest any precise questions; "So?", after some statements is not much of a question. Ask better questions, and get quicker and better replies.
-
I read your post a few days ago, but it was a bit tangential to what I thought we were discussing, and your language really didn't suggest any precise questions; "So?", after some statements is not much of a question. Ask better questions, and get quicker and better replies.
[a href=\"index.php?act=findpost&pid=179283\"][{POST_SNAPBACK}][/a]
I'm referring to the images I linked to above. They are very interesting as we can precisely compare the performance of different cameras. There were some blown up highlights in the F30 output (e.g. white thread), and nice detail in the dark tones, e.g. in the black napkin.
This detail is absent from the Fuji F50 image (10 MP) and the Canon SD950 (12 MP). The patterns on the napkin are close to invisible.
Just explain me why Fuji engineers were not able to rescue decent signal in the dark tones with the 10 MP sensor while they could with the 6.3 MP one. OK, on the F50 they reduced the clipping in the highlights, but this could have been done while still getting as much signal in the shades (and only reducing contrast somewhere, for example in the shades...). In contrast (no pun intended) the Canon produces loads of noise in the black napkin, and the Fuji has erased all detail there.
If your oversampling theory is correct, the performance of the 10/12 MP should be at least on par with the 6.3 MP on these dark patterns.
My "so?" means: now up to you to try and justify your point that more pixels is better. I don't have access to the raw files, but we'll suppose that what Fuji electronics were able to do on them two years ago, they're still able to do now.
P.
PS: and please don't tell me the apparent F30 superiority at ISO 800 is all down to overaggressive NR. There's (far) less NR in the Canon image and it's pretty clear from that image that more NR would have cleared all the white patterns on the black napkin: that's actually what happens on the F50. In the less heavily processed Canon JPEGs the patterns are swamped by noise, and already close to invisible.
PPS: Look also at the reflection on the Wychwood ale bottle. On the F30 you can discern their setup between the softboxes, you can't on the Canon or F50. There's just a patch of gray divided by a fine black line. And by the way it's close to the center so lens quality differences should not matter much.
-
I'm referring to the images I linked to above. They are very interesting as we can precisely compare the performance of different cameras.
IMO, cameras that only outputs JPEGs that have extreme noise reduction are cartoon machines, not cameras.
There were some blown up highlights in the F30 output (e.g. white thread), and nice detail in the dark tones, e.g. in the black napkin.
That's a clue to you right there; the F30 received better exposure, so naturally, the shadows are going to have less noise, and fall into a higher level with lower noise reduction.
This detail is absent from the Fuji F50 image (10 MP) and the Canon SD950 (12 MP). The patterns on the napkin are close to invisible.
Just explain me why Fuji engineers were not able to rescue decent signal in the dark tones with the 10 MP sensor while they could with the 6.3 MP one. OK, on the F50 they reduced the clipping in the highlights, but this could have been done while still getting as much signal in the shades (and only reducing contrast somewhere, for example in the shades...).
No, it couldn't have been done that way. If one is clipping and the other one isn't, then they received different relative (and probably absolute) exposures. This is not an engineering issue per se; it is a programmer's issue, as the noise reduction (not just the noise) hides detail.
In contrast (no pun intended) the Canon produces loads of noise in the black napkin, and the Fuji has erased all detail there.
If your oversampling theory is correct, the performance of the 10/12 MP should be at least on par with the 6.3 MP on these dark patterns.
Through this wreckage of noise reduction? You must be kidding.
The fact is, every bright area on the F50 is clearly superior to the F30 in terms of resolution. I can clearly read things on the labels in the F50 image that I struggle with or can't read at all in the F30 image. Everything that isn't below a certain exposure level where the Fuji NR tries to remove all detail is more detailed in the F50.
My "so?" means: now up to you to try and justify your point that more pixels is better. I don't have access to the raw files, but we'll suppose that what Fuji electronics were able to do on them two years ago, they're still able to do now.
We're looking at software as much as we are looking at electronics, and apparently through unequal exposure.
P.
PS: and please don't tell me the apparent F30 superiority at ISO 800 is all down to overaggressive NR. There's (far) less NR in the Canon image and it's pretty clear from that image that more NR would have cleared all the white patterns on the black napkin: that's actually what happens on the F50. In the less heavily processed Canon JPEGs the patterns are swamped by noise, and already close to invisible.
Less heavily processed? You must be kidding. Have you ever, in your life, seen how much real noise is in RAW captures from compact's sensors? The Canon is extremely noise-reduced too; it just doesn't have an edge-preserve priority like Fuji does. *EVERY* compact sensor has chromatic noise when converted literally, even in the midtones at ISO 100.
PPS: Look also at the reflection on the Wychwood ale bottle. On the F30 you can discern their setup between the softboxes, you can't on the Canon or F50. There's just a patch of gray divided by a fine black line. And by the way it's close to the center so lens quality differences should not matter much.
[a href=\"index.php?act=findpost&pid=179337\"][{POST_SNAPBACK}][/a]
I don't see any difference in the Fujis. You're going to have to be more specific. I'm not even going to look at the Canon. The Canon is low-pass-filtered into oblivion. I will never, ever again buy a small camera that doesn't do RAW.
-
This has been a very interesting thread. I am very familiar with the power of oversampling in practice. That said, I have never been able to get high ISO images from any small sensor/high MP camera to have as much detail/noise as my old F30 (now sold) when considered at equal print sizes. I can get close, but not all the way there. I have seen John's example(s) of S6500 high ISO RAW images (same as F30 RAW), and I am unconvinced that on-board NR is the answer to the F30 capabilities. As hellish as those RAWs look, they look better than downsampled images from high MP small sensor cameras. I have tried RAW files from many high-end 10MP+ compacts, and after downsampling to 6MP with application of NR (Noise Ninja, Neat Image, Dfine, etc, I have tried many) at any step in the process, I cannot match the in-camera F30 result. I can get much closer to that result than one might guess based on a misguided pixel-level review prior to downsampling, but I can't get all the way there. Same goes for downsampling high ISO (3200+), NR-applied 1Ds III files to 12MP and comparing them to D3 files. The D3 files still win, in terms of final detail/noise at an equal print/viewing size, in every comparison I have seen. In the end, I understand why the high MP cameras should be able to match the result, but in practice I don't see it happening. I'm guessing that John and others believe that it comes down to processing software and skills. I'll believe that when I see a good, practical demonstration of it.
With regards to aliasing in the DP1 images, it is definitely there. How bothersome one finds it is a personal issue. All aspects including aliasing artifacts being considered, the DP1 samples I have seen are more pleasing to me, at moderate print sizes, than those of any other compact camera I have yet seen.
-
This has been a very interesting thread. I am very familiar with the power of oversampling in practice. That said, I have never been able to get high ISO images from any small sensor/high MP camera to have as much detail/noise as my old F30 (now sold) when considered at equal print sizes. I can get close, but not all the way there. I have seen John's example(s) of S6500 high ISO RAW images (same as F30 RAW), and I am unconvinced that on-board NR is the answer to the F30 capabilities. As hellish as those RAWs look, they look better than downsampled images from high MP small sensor cameras. I have tried RAW files from many high-end 10MP+ compacts, and after downsampling to 6MP with application of NR (Noise Ninja, Neat Image, Dfine, etc, I have tried many) at any step in the process, I cannot match the in-camera F30 result. I can get much closer to that result than one might guess based on a misguided pixel-level review prior to downsampling, but I can't get all the way there. Same goes for downsampling high ISO (3200+), NR-applied 1Ds III files to 12MP and comparing them to D3 files. The D3 files still win, in terms of final detail/noise at an equal print/viewing size, in every comparison I have seen. In the end, I understand why the high MP cameras should be able to match the result, but in practice I don't see it happening. I'm guessing that John and others believe that it comes down to processing software and skills. I'll believe that when I see a good, practical demonstration of it.
With regards to aliasing in the DP1 images, it is definitely there. How bothersome one finds it is a personal issue. All aspects including aliasing artifacts being considered, the DP1 samples I have seen are more pleasing to me, at moderate print sizes, than those of any other compact camera I have yet seen.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=180614\")
I used to use a Fuji f10, then a Fuji f50fd, and now a Ricoh gx100.
I really miss my fuji f10. then f50fd was pretty horrible at low light.
I'm actually preparing to get another fuji f10 to do my work
All this talk about technical aspects sometimes overlooks a very basic emotional connection on : how does this picture look to me?
I've tried different cameras from Canon to Panasonic to Ricoh to Fuji and I have to say that the Fuji F10 produced for me a very photographic texture in its images which I managed to interpolate up to 33x44 inches with beautiful results and 8ft x 11ft with acceptable results (look here: [a href=\"http://superhyperreal.com/test/foodp2.html]http://superhyperreal.com/test/foodp2.html[/url] in print http://superhyperreal.com/NEWS/fcb6.jpg (http://superhyperreal.com/NEWS/fcb6.jpg))
People always talk about getting the effect later in post production but I treat cameras like lenses nowadays (different brands give different looks)
and frankly different cameras give different emotional connections (sometimes you just feel that using a certain camera for a certain type of work feels more..right. Same thing as you view the image on a screen)
I use my Ricoh for alot of general shots but I am looking to reuse my Fuji for my food shots because I can't get the "feeling" in the shots I am looking from my Ricoh for the food shots.
Conversely, you will see alot of people raving about their ricohs for certain kinds of work (street shooting, b/w) even though the IQ of the Ricoh is probably not the best.
Technical debate is great but this is just a little hello from the other side.