Pages: 1 ... 5 6 [7] 8 9 ... 11   Go Down

Author Topic: 1Ds MKIII and Optical Low Pass filtering  (Read 149534 times)

juicy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 254
1Ds MKIII and Optical Low Pass filtering
« Reply #120 on: October 31, 2007, 05:25:13 am »

Quote
Those photographers who have invested in the new Digitar and Rodenstock lenses which have a reduced image circle more appropriate for the 'cropped' MF format of the P22, P25 and P45 must be hopping mad at the amount of aliasing they are getting. Some of those new MF lenses have an MTF response higher that that of any 35mm lens that I've seen. Their performance looks more like that of the Zuiko lenses for the 4/3rds format, you know, something like 70% MTF at 60 lp/mm (can't remember off-hand the precise figures).
[{POST_SNAPBACK}][/a]


Hi!

In some occasions there are moire problems with digital backs and not only with Schneider and Rodenstock (Linos) digital-optimized lenses. Software is getting better in dealing with moire in PP.
Two different moire examples can be found in [a href=\"http://luminous-landscape.com/forum/index.php?showtopic=13563]http://luminous-landscape.com/forum/index....showtopic=13563[/url]  
 (have a look at the links in the first post, 33mpix with different lenses) and http://luminous-landscape.com/forum/index....showtopic=12976  (22mpix).
I want to emphasize that I don't wish anyone to draw any conclusions about these particular backs or the photographers work based on these couple of pictures, these are only to demonstrate the different moire effects encountered in different situations.

Cheers,
J
Logged

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
1Ds MKIII and Optical Low Pass filtering
« Reply #121 on: October 31, 2007, 06:41:18 am »

Quote
Those photographers who have invested in the new Digitar and Rodenstock lenses which have a reduced image circle more appropriate for the 'cropped' MF format of the P22, P25 and P45 must be hopping mad at the amount of aliasing they are getting.

I don't think that's necessarily the case, at least most of the time. Improved RAW interpolation algorithms and software tools are reducing the likelihood of color aliasing being a problem even with troublesome subjects. And luminance moire is less unnatural looking when part of a ~40MP image than in a ~4MP image, at least at viewing distances where one is looking at the entire image.
Logged

Lin Evans

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 269
    • http://www.lin-evans.net
1Ds MKIII and Optical Low Pass filtering
« Reply #122 on: October 31, 2007, 09:54:24 am »

Time to put up or shut up.

Let's see some Foveon images with all this "nasty aliasing" you've been peddling. I've been listening to you and John and Graeme wax eloquent about all the problems with Foveon images and not a hint of anything except hot air from all of you.

My Foveon images and numerous others hang at PMA in plain sight at greater than A0 size and not a single comment from the thousands of visitors about "nasty aliasing". In fact the vast preponderance of comments have been exceedingly complementary. I can show you a 2.5 gigapixel stitched image made with an SD14 which prints beautifully at "any" size. So how about the three of you get together and show all your results with a Foveon processor which substantiate your continuing BS??

Lin

Quote
I don't think that's necessarily the case, at least most of the time. Improved RAW interpolation algorithms and software tools are reducing the likelihood of color aliasing being a problem even with troublesome subjects. And luminance moire is less unnatural looking when part of a ~40MP image than in a ~4MP image, at least at viewing distances where one is looking at the entire image.
[a href=\"index.php?act=findpost&pid=149757\"][{POST_SNAPBACK}][/a]
Logged
Lin

Dave Millier

  • Full Member
  • ***
  • Offline Offline
  • Posts: 146
    • Whispering Cat Photography
1Ds MKIII and Optical Low Pass filtering
« Reply #123 on: October 31, 2007, 01:39:00 pm »

Hi Folks

I can see this discussion about the Foveon sensor performance has ruffled a few feathers. I'm a veteran of this, being reknowned on the sigma forum of "another place" as being somewhat more cautious than is typical for that forum.

I'd like to add one or two thoughts of my own...

I recently purchased a used Sd9 in order to do my own testing of the Foveon concept (and perhaps to improve my credibilty in the endless debates).

My favourite camera is the 350D (mainly for the compact size and decent results) so I was very interested to see how the Sd9 compared to that camera. In a nutshell, the answer is pretty well.  I found no discernable detail difference between the two cameras even pixel peeping at 100% and certainly not in prints.

There is a difference in the way aliasing is handled though: the Canon shows the usual "maze" type artifacts and moire while the SD9 tends to reproduce shrinking regular patterns as a smaller number of larger blocks at the resolution limit. Neither is ideal, but is really a matter of taste as to which is least bad.

Here's a link to a 100% crop comparison that shows the similarity in detail and the difference in aliasing: http://forums.dpreview.com/forums/read.asp...essage=23492002

In addition to the Sd9 v 350D shootout, I've been working with a Kodak 14nx owner who recently acquired a Sd14 to test how these cameras compare for resolution.  I've been delayed in writing up the exercise because of other committements but as a quick summary: it is impossible to tell which camera produced the prints in prints up to 24x16 inches.

I'm not sure where this fits with the general debate but I've no doubt the SD14 is capable of producing prints every bit as good as a camera like the 5D - up to this print size at least.

I think at the end of the day it is important to reference all debates to physical evidence.  Hopefully, I can get my act together and actually publish my evidence soon, but in the meantime, can I suggest that the debate here would be moved forward more efficiently if arguments were supported by some comparison images from real world photography?

Regards

Dave Millier
Logged
My website and photo galleries: http://w

CatOne

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 458
    • http://blloyd.smugmug.com
1Ds MKIII and Optical Low Pass filtering
« Reply #124 on: October 31, 2007, 01:50:52 pm »

You need to stop bringing sensibility into the debate.  We have analytical preconceived notions to argue about.

 

I think the only reason this thread still persists is because the shopkeeper is away for a few weeks ;-)
Logged

djgarcia

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 341
    • http://improbablystructuredlayers.net
1Ds MKIII and Optical Low Pass filtering
« Reply #125 on: October 31, 2007, 01:56:14 pm »

Hey, let's not forget the subjective entertainment value to help us wile away those boring work hours  ...
Logged
Over-Equipped Snapshooter - EOS 1dsII &

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
1Ds MKIII and Optical Low Pass filtering
« Reply #126 on: October 31, 2007, 04:04:13 pm »

Quote
So how about the three of you get together and show all your results with a Foveon processor which substantiate your continuing BS??

Are you referring to the BS where I agree that Foveon images are better per-pixel than Bayer? Or the BS where I point out that it is mathematically impossible for any sampled system to record an accurate reproduction of the input above Nyquist, which you seem to be claiming with your defense of the test results showing Foveon resolving more lines of resolution than it actually has?

Contrary to your claims, I am NOT anti-Foveon. Capturing all 3 primaries at each pixel site is better than capturing only one. But every comparison I have made and seen indicates that the advantages of Foveon, while real, are significantly exaggerated by the more enthusiastic proponents of the technology such as yourself. I have never seen any evidence to support the notion that Foveon pixels are twice as good as Bayer--4MP Foveon = 8MP Bayer in overall image quality and detail--my analysis indicates that 3MP Foveon = ~5MP Bayer when both are optimally sharpened. That is a solid performance advantage for the Foveon, at least on a per-pixel basis.

Foveon's crippling weakness right now is its low pixel count. No matter how good your ~4MP are, there's only so much detail you can cram into those pixels. Here's an experiment anyone can try at home to demonstrate this. Start out with a 1Ds RAW, and open it in Photoshop in 16-bit mode at its original resolution (4064x2704). Make a copy of it. Now downsize the copy using Bicubic Sharper, and print both files the same size (say 12x18 inches) and compare the prints. Undo the changes, and repeat the resizing, reducing by a greater percentage each time. How small can you resize the file before significant loss of image quality and detail starts showing up in the print?

When you size the image down to 60% of the original pixel count (77.46% of original linear dimensions) the damage to the image is very slight; one has fewer pixels, but they are better quality, and there is little difference to be seen in prints. Reducing to 50% of the pixel count (70.71% of original linear dimensions), one starts seeing a loss of fine detail. The individual pixels in the resized image are better than in the original image, but there aren't enough of them to keep the high-frequency detail and the image as a whole suffers. It follows then that given a perfect sensor that records all color channels for each pixel, one can achieve similar image quality with somewhere between 50-60% of the pixel count of a comparable Bayer sensor. This notion tracks fairly well with the comparisons I've done between the 10D and SD9; Foveon approaches, but does not quite match the IQ and detail of a Bayer sensor with double the pixel count.

If the most exuberant claims about Foveon were true (X Foveon photosites = X Bayer photosites), then you could reduce any Bayer-sensor image to 57.74% of its original pixel dimensions (which will leave you with 1/3 of the original pixel count) with little or no loss of image quality or detail when printed. But this is most certainly not the case; reducing 4064x2704 pixels down to 2347x1561 tosses out an awful lot of fine detail. The remaining pixels are very nice on a per-pixel basis, but the image as a whole has suffered irreparable damage; much of the fine detail present in the original capture is gone. There are simply not enough pixels to hold all of it. So the most extreme claims about Foveon are simply not credible.

I would LOVE to see a Foveon sensor with a pixel count that is more competitive with Bayer-sensor cameras. But given the current pixel count limitations of available Foveon sensors, I can do better with my 1Ds.
« Last Edit: October 31, 2007, 04:49:26 pm by Jonathan Wienke »
Logged

Graeme Nattress

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 584
    • http://www.nattress.com
1Ds MKIII and Optical Low Pass filtering
« Reply #127 on: October 31, 2007, 04:10:47 pm »

Quote
the Canon shows the usual "maze" type artifacts and moire while the SD9 tends to reproduce shrinking regular patterns as a smaller number of larger blocks at the resolution limit. Neither is ideal, but is really a matter of taste as to which is least bad. [a href=\"index.php?act=findpost&pid=149831\"][{POST_SNAPBACK}][/a]

Maze artifacts come from poor demosaic artifacts. The demosaicing I'm using the the RED camera will never produce such nasties! I don't know why people code demosaics that bad....

Graeme
Logged

Dave Millier

  • Full Member
  • ***
  • Offline Offline
  • Posts: 146
    • Whispering Cat Photography
1Ds MKIII and Optical Low Pass filtering
« Reply #128 on: October 31, 2007, 04:23:28 pm »

I have no intention of disputing your numbers but it is conceivable that some bayer sensors are worse than you might expect, therefore making the Foveon seem a better performer.

As always, platonic theories must be tempered by boring facts. In my previous post, I provided some explicit evidence that the SD9 can compete with the 8MP 350D - something that surprised me.  There appear to be only two explanations: either the Foveon is better than you (and me) expected or the Canon sensor is worse.  Do you have any comment on this?
 
 
Quote
Are you referring to the BS where I agree that Foveon images are better per-pixel than Bayer? Or the BS where I point out that it is mathematically impossible for any sampled system to record an accurate reproduction of the input above Nyquist, which you seem to be claiming with your defense of the test results showing Foveon resolving more lines of resolution than it actually has?

[snip]

I have never seen any evidence to support the notion that Foveon pixels are twice as good as Bayer--4MP Foveon = 8MP Bayer in overall image quality and detail--my analysis indicates a bit less than 50% advantage, so that 4MP Foveon = ~5.5MP Bayer when both are optimally sharpened. That is a solid performance advantage for the Foveon, at least on a per-pixel basis.

[a href=\"index.php?act=findpost&pid=149856\"][{POST_SNAPBACK}][/a]
Logged
My website and photo galleries: http://w

Lin Evans

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 269
    • http://www.lin-evans.net
1Ds MKIII and Optical Low Pass filtering
« Reply #129 on: October 31, 2007, 04:41:07 pm »

Quote
Are you referring to the BS where I agree that Foveon images are better per-pixel than Bayer? Or the BS where I point out that it is mathematically impossible for any sampled system to record an accurate reproduction of the input above Nyquist, which you seem to be claiming with your defense of the test results showing Foveon resolving more lines of resolution than it actually has?

Let me quote myself from a previous ansewer to Ray, and you can explain which part of this rather plain English you fail to understand............

-------------------------------------------
.........The 10.3 million Foveon processor used in the SD9/SD10 only produces a measured count of 1550 lines horizontal/vertical (in reality, it can't be more than the vertical display pixel count of 1512 lines) but human visual acuity and standardized resolution chart print error account for the variance as observed by testing (Phil Askey on dPReview's figures). The same margin for error applies to all other cameras tested so if I accept their numbers then I must also accept the "measured" Foveon numbers of 1550 lines vertical and horizontal. The "measured" color resolution is identical to the b&w resolution for this processor. ..............
----------------------------------------------

Contrary to your claims, I am NOT anti-Foveon. Capturing all 3 primaries at each pixel site is better than capturing only one. But every comparison I have made and seen indicates that the advantages of Foveon, while real, are significantly exaggerated by the more enthusiastic proponents of the technology such as yourself. I have never seen any evidence to support the notion that Foveon pixels are twice as good as Bayer--4MP Foveon = 8MP Bayer in overall image quality and detail--my analysis indicates a bit less than 50% advantage, so that 4MP Foveon = ~5.5MP Bayer when both are optimally sharpened. That is a solid performance advantage for the Foveon, at least on a per-pixel basis.

Did I say you were anti-Foveon? When and where did I say this?? If you would spend more time reading and understanding and less time with knee jerk reactions based on what you apparently "want" to believe you would know that what I said was that the "nasty aliasing" issue is BS.

Foveon's crippling weakness right now is its low pixel count. No matter how good your ~4MP are, there's only so much detail you can cram into those pixels. Here's an experiment anyone can try at home to demonstrate this. Start out with a 1Ds RAW, and open it in Photoshop in 16-bit mode at its original resolution (4064x2704). Make a copy of it. Now downsize the copy using Bicubic Sharper, and print both files the same size (say 12x18 inches) and compare the prints. Undo the changes, and repeat the resizing, reducing by a greater percentage each time. How small can you resize the file before significant loss of image quality and detail starts showing up in the print?

And what I'm telling you ois that Foveon doesn't have a "crippling weakness". Forget your "experiments" and look at the images and the huge prints at PMA and Photokina. Your entire argument is predicated on your experience with CFA cameras and your incorrect assumptions about Foveon "weakness". I'm telling you that people who use the cameras on a daily basis including CFA cameras and including the 1DS and 5D and D2X don't see a "crippling weakness".  People just as skilled or more skilled than you with pixel peeping such as the author of Qimage (Mike Chaney) have put aside their full frame Canons and favor the 1DS for precisely IMAGE QUALITY.

When you size the image down to 60% of the original pixel count (77.46% of original linear dimensions) the damage to the image is very slight; one has fewer pixels, but they are better quality, and there is little difference to be seen in prints. Reducing to 50% of the pixel count (70.71% of original linear dimensions), one starts seeing a loss of fine detail. The individual pixels in the resized image are better than in the original image, but there aren't enough of them to keep the high-frequency detail and the image as a whole suffers. It follows then that given a perfect sensor that records all color channels for each pixel, one can achieve similar image quality with somewhere between 50-60% of the pixel count of a comparable Bayer sensor. This notion tracks fairly well with the comparisons I've done between the 10D and SD9; Foveon approaches, but does not quite match the IQ and detail of a Bayer sensor with double the pixel count.

Working with Foveon images and working with CFA images are totally different issues. Foveon not only matches the IQ and detail of a Bayer sensor with double the file size in pixels, the 3.4 megapixel Foveon exceeds the measured detail of the 6.3 mp Bayer and gives better IQ per the subjective evaluation of many, many photographers.


If the most exuberant claims about Foveon were true (X Foveon photosites = X Bayer photosites), then you could reduce any Bayer-sensor image to 57.74% of its original pixel dimensions (which will leave you with 1/3 of the original pixel count) with little or no loss of image quality or detail when printed. But this is most certainly not the case; reducing 4064x2704 pixels down to 2347x1561 tosses out an awful lot of fine detail. The remaining pixels are very nice on a per-pixel basis, but the image as a whole has suffered irreparable damage; much of the fine detail present in the original capture is gone. There are simply not enough pixels to hold all of it. So the most extreme claims about Foveon are simply not credible.

I don't know what the "most extreme claims" about Foveon are and really could care less. What I do know very well is real-world "results" based on personally inspecting many, many thousand Sigma prints and many hundred thousand CFA prints.


I would LOVE to see a Foveon sensor with a pixel count that is more competitive with Bayer-sensor cameras. But given the current pixel count limitations of available Foveon sensors, I can do better with my 1Ds.

That's "your" opinion. My opinion is that I can't and it's an opinion shared by many who have real-world experience with both.

Lin
[a href=\"index.php?act=findpost&pid=149856\"][{POST_SNAPBACK}][/a]
« Last Edit: October 31, 2007, 04:41:38 pm by Lin Evans »
Logged
Lin

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
1Ds MKIII and Optical Low Pass filtering
« Reply #130 on: October 31, 2007, 05:07:58 pm »

Quote
Foveon not only matches the IQ and detail of a Bayer sensor with double the file size in pixels, the 3.4 megapixel Foveon exceeds the measured detail of the 6.3 mp Bayer

Yes, when you buy into invalid measurement methodology that gives a higher resolution rating than the sensor has line pairs. That claim is no different than saying that a CD with a 44.1KHz sampling rate can meaningfully record a 25KHz audio signal.

As to the subjective opinion of professional photographers, I think its safe to say that a lot more pros are shooting with Canon DSLRs than Sigma, most because they believe they are getting better image quality with Canon. Sigmas aren't the only DSLR capable of capturing a large file that can print large with high quality.
Logged

Lin Evans

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 269
    • http://www.lin-evans.net
1Ds MKIII and Optical Low Pass filtering
« Reply #131 on: October 31, 2007, 05:46:49 pm »

You still can't read??? Still attributing things to me which I never said. Don't you get tired of being wrong?  So what's your "valid measurement methodology?" Maybe you can inform the rest of the world and camera reviewers will get results "you" like - LOL. Perhaps if they keep testing they will find a way to substantiate measurements which you agree with. Strange thing is that when using the identical approach for CFA and Foveon they all seem to come up with similar results, non of which seem to agree with your assessment. Imatest analysis (oops - know issues causing the significantly better Foveon results viz 6mp Bayer CFA, hmmm), plain old b&w standardized resolution chart photos, color resolution chart photos, all seem to validate better Foveon results. Eye witnesses with prints in hand saying no difference between enlargements with 3.4 megapixel Foveon and 8 megapixel CFA - hmmm.

Quote
Yes, when you buy into invalid measurement methodology that gives a higher resolution rating than the sensor has line pairs. That claim is no different than saying that a CD with a 44.1KHz sampling rate can meaningfully record a 25KHz audio signal.

As to the subjective opinion of professional photographers, I think its safe to say that a lot more pros are shooting with Canon DSLRs than Sigma, most because they believe they are getting better image quality with Canon. Sigmas aren't the only DSLR capable of capturing a large file that can print large with high quality.
[a href=\"index.php?act=findpost&pid=149875\"][{POST_SNAPBACK}][/a]

I think it's safe to say that there are a lot more Canon, Nikon, Olympus, Pentax, Sony, Fuji and Kodak dSLR's than Sigma. I think it's also safe to say that the majority of professionals have a rather large investment in lenses and are unlikely to jump from one brand to another unless they are not getting results which are satisfactory. I get great results from a number of different dSLR's, fixed lens digicams and platforms and that includes Sigma. No "nasty artifacts" - no problems with "crippled" resolution or poor image quality.

I've yet to have a customer ask me which type of "camera" I used to produce an image. They either like it or they don't. The majority of the time they do or I wouldn't have had a successful career and enough income to buy all the equipment I need or want.

It's obvious we are never going to agree about this so not much sense continuing. You love numbers and tests - I love "results". I can't sell numbers and tests but I sure can sell results.

Yes, it's possible to get great large prints from CFA cameras as well. My D2XS produces incredible quality as does my SD14 and SD10 and my 1DS. They each have strengths and weaknesses. To my knowledge I've never said othewise. What I have said is that my SD14 produces, in my opinion, the best IQ of all for many of my purposes. I much prefer it for landscapes. It doesn't produce "nasty artifacts" any more than my CFA cameras do. It gives up nothing in IQ to any of my CFA cameras for subjects and under conditions it's amenable to. It's not a good low light, high ISO camera. It's not very useful for shooting birds in flight or for shooting sports. Horses for courses.....

Lin

Logged
Lin

Quentin

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1222
    • Quentin on Facebook
1Ds MKIII and Optical Low Pass filtering
« Reply #132 on: October 31, 2007, 05:59:48 pm »

Whatever happened to the Sigma compact?  That was a camera I would have been and might still be interested in, a good way to test the Foveon sensor out.

And why are there no higher pixel count sensors from foveon?  Seems a shame we don't have a 10mp (x3 if you like) foveon based dslr to choose from.

Quentin
Logged
Quentin Bargate, ARPS, Author, Arbitrato

Dave Millier

  • Full Member
  • ***
  • Offline Offline
  • Posts: 146
    • Whispering Cat Photography
1Ds MKIII and Optical Low Pass filtering
« Reply #133 on: October 31, 2007, 06:08:38 pm »

Jonathan

I believe it was the host of this very site who published an article called "Cognitive Dissonance"...

I have crossed swords with Lin many times over Foveon but once again it all comes down to evidence.  I've posted a comparison between 3.4MP X3 and 8MP bayer. It's not a lie - it shows the SD9 really does match the Canon for detail in every practical sense and this holds true for every base ISO comparison I've made.  

I'd value your opinion of the evidence rather than the theory. Do you see anything in those crops that surprises you; do you have an alternative explanation of how the Foveon does so well?

I don't personally believe that the Foveon sensor is better than 2x more efficient than Bayer sensors in an absolute sense. My "theory" is that the 8MP canon sensor isn't really very much more detailed than its 6MP predecessors and that is the likely explanation of why the Foveon holds up. Whatever the reason, there is a piece of evidence that requires a better analysis than blindly quoting numbers...

 

Quote
Yes, when you buy into invalid measurement methodology that gives a higher resolution rating than the sensor has line pairs. That claim is no different than saying that a CD with a 44.1KHz sampling rate can meaningfully record a 25KHz audio signal.

As to the subjective opinion of professional photographers, I think its safe to say that a lot more pros are shooting with Canon DSLRs than Sigma, most because they believe they are getting better image quality with Canon. Sigmas aren't the only DSLR capable of capturing a large file that can print large with high quality.
[a href=\"index.php?act=findpost&pid=149875\"][{POST_SNAPBACK}][/a]
Logged
My website and photo galleries: http://w

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10316
1Ds MKIII and Optical Low Pass filtering
« Reply #134 on: November 01, 2007, 10:02:45 am »

Quote
I have crossed swords with Lin many times over Foveon but once again it all comes down to evidence.  I've posted a comparison between 3.4MP X3 and 8MP bayer. It's not a lie - it shows the SD9 really does match the Canon for detail in every practical sense and this holds true for every base ISO comparison I've made. 
[a href=\"index.php?act=findpost&pid=149892\"][{POST_SNAPBACK}][/a]

I agree. One always needs the evidence otherwise it's not clear what people are referring to.

I've looked at your 100% crops comparing the 350D with the SD9 and I see relative advantages and disadvantages to both images.

To my eyes the SD9 image looks marginally sharper all over. Not in the sense of providing more detail but in the sense of greater accutance, as though one were comparing a prime lens with a zoom lens. I'm not sure if this is due to greater sharpening of the SD9 image, however.

The balustrade and the red tiled roof tend to highlight the SD9 aliasing problems. The vertical struts in each section of the balustrade are at least identifiable in the 350D image, although the number of struts visible in each section steadily diminishes as one moves from far right to left. In the SD9 image, the struts are clearly visible and countable only in the far right segment. Strong aliasing is very noticeable as one moves from right to left and before long everything is complete mush and color artifacts with no countable struts visible at all.

Nevertheless, I can see why some people would prefer the SD9 image and others the 350D image.

Thanks for providing the comparison images, Dave.
Logged

Lin Evans

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 269
    • http://www.lin-evans.net
1Ds MKIII and Optical Low Pass filtering
« Reply #135 on: November 01, 2007, 01:48:02 pm »

Hi Ray,

You have them backward. The top image is Sigma, the bottom Canon. The mush and color aliasing are on the Canon image. The Sigma image still shows the struts as identifiable only fewer than are actually there up until the sixth group segment.

The Sigma won't produce color moire. Jonathan might learn something by looking at this example of color moire he has never seen in 120,000 frames. Also it may be a learning experience for John and his "Tetris" pieces theory. The number of struts in the Sigma image diminish and are not correct, but they are still identifiable as "struts" very far past Nyquist. Most human observers looking only at the aliased struts when asked what they "thought" they were seeing would properly identify these as "struts" where seen in isolation in the Canon image it would be difficult to say unless the image was desaturated. Likewise with grass, pine needles, leaves, etc., which is one reason many like the Sigma landscapes. They don't see blobs of indiscriminate color and shapelessness in their landscape enlargements. They may see aliasing but the aliasing looks a lot better and much more closely resembles "reality" to the human brain than what is there with an AA filtered Bayer image. Of course one could always argue that an image shouldn't be enlarged to the degree that this color moire and shapeless blobs become visible - that's an issue for a different discussion.

Best regards,

Lin

Quote
I agree. One always needs the evidence otherwise it's not clear what people are referring to.

I've looked at your 100% crops comparing the 350D with the SD9 and I see relative advantages and disadvantages to both images.

To my eyes the SD9 image looks marginally sharper all over. Not in the sense of providing more detail but in the sense of greater accutance, as though one were comparing a prime lens with a zoom lens. I'm not sure if this is due to greater sharpening of the SD9 image, however.

The balustrade and the red tiled roof tend to highlight the SD9 aliasing problems. The vertical struts in each section of the balustrade are at least identifiable in the 350D image, although the number of struts visible in each section steadily diminishes as one moves from far right to left. In the SD9 image, the struts are clearly visible and countable only in the far right segment. Strong aliasing is very noticeable as one moves from right to left and before long everything is complete mush and color artifacts with no countable struts visible at all.

Nevertheless, I can see why some people would prefer the SD9 image and others the 350D image.

Thanks for providing the comparison images, Dave.
[a href=\"index.php?act=findpost&pid=150034\"][{POST_SNAPBACK}][/a]
« Last Edit: November 01, 2007, 02:22:34 pm by Lin Evans »
Logged
Lin

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10316
1Ds MKIII and Optical Low Pass filtering
« Reply #136 on: November 02, 2007, 03:11:27 am »

Quote
Hi Ray,

You have them backward. The top image is Sigma, the bottom Canon. The mush and color aliasing are on the Canon image. The Sigma image still shows the struts as identifiable only fewer than are actually there up until the sixth group segment.

[a href=\"index.php?act=findpost&pid=150085\"][{POST_SNAPBACK}][/a]

LOL! Well, there you are then. I could have sworn the bottom image, which shows obvious aliasing, was from the SD9. Considering the SD9 is 3.4mp, it would not be difficult to convince me, had the SD14 been used in the comparison, that the Foveon image would have been preferred to the 8mp Bayer type in all respects... at least in these examples. One would need to compare many different types of images to be certain about this though.

The fact that the 350D image appears to have greater accutance (like comparing a prime with a zoom) is probably attributable to its additional 2mp. The SD9 was always considered to be on a par with a 6mp Bayer type.
Logged

juicy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 254
1Ds MKIII and Optical Low Pass filtering
« Reply #137 on: November 02, 2007, 04:56:23 am »

Hi!

The lower image from 350D shows a very typical Canon color aliasing. With a sharp lens (and even the WA zooms are sharp enough) there is this problem when photographing subjects with regular high frequency patterns such as fabrics, buildings etc. I have encountered this numerous times with several Canons. Digital backs aren't immune either. Software is getting better in dealing with this (Hasselblad's Phocus is advertised as particularly effective in this) but in most cases it means less than perfect results and more PP.

Cheers,
J
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1Ds MKIII and Optical Low Pass filtering
« Reply #138 on: November 02, 2007, 06:03:33 am »

Quote
The fact that the 350D image appears to have greater accutance (like comparing a prime with a zoom) is probably attributable to its additional 2mp. The SD9 was always considered to be on a par with a 6mp Bayer type.
[a href=\"index.php?act=findpost&pid=150192\"][{POST_SNAPBACK}][/a]

He said these are both 100% crops, which is a disadvantage to the 350D, as a smaller fraction of the image is used, which always results in lower quality.

The only fair way to compare on a monitor is to use the same FOV lens with both cameras, and upsample both so that they have the subject size, and step back and view from a distance, if necessary.  Even when printing small, images are resampled, as the brain-dead drivers resample before printing, so you must print large for a fair comparison.
Logged

Dave Millier

  • Full Member
  • ***
  • Offline Offline
  • Posts: 146
    • Whispering Cat Photography
1Ds MKIII and Optical Low Pass filtering
« Reply #139 on: November 02, 2007, 06:15:14 am »

Hi Ray

I don't think there is any difference in detail. Acutance is down to sharpening differences. Straight out of the raw converter, the SD9 is much, much sharper - the Canon looks typically soft. After, uprezzing to match the Canon file size both were sharpened. I don't see much of a difference but this can be fine tuned to taste and application.

The main difference is the aliasing.  Both cameras alias but in different ways. That balustrade is a beautiful test chart.  Essentially what you are looking at is colour aliasing vs luminance aliasing. The canon shot is contaminated by false colour (despite the presence of the AA filter), the SD9 shows no colour artifacts. But you do get false detail.

In many ways this false detail lies at the heart of the debate. Is is good, bad or neutral - it clearly isn't "correct". You see the good and the bad in this shot.

The balustrade shows what happens really neatly. Essentially as the detail shrinks below Nyquist, it stops being reproduced correctly. Instead, you get false detail but the false detail is related to the real detail in a way that is useful at creating the impression of real detail.  In a print of this shot what the eye sees is a convincing impression of higher resolution than the 350D (even though technically the detail is artifacts).

The downside is the tiled roof of the boathouse. This is a standard horizontal tiling patterm yet the SD9 reproduces it as a strange diagonal striped moire effect. The 350D does badly here too: if you blow this shot up it become evident that it has produced a patchy effect. Some areas show tiles, some a blurred smudge and some a kind of maze/hatched effect. Fortunately for the 350D, the whole thing is smudged enough to mix these effects together into something that looks passable.

And there I think lies the rub: which is best (or least worse) depends on subject, angle, magnification, personal taste. And we know exactly how this kind of ambiguity leads to squabbling.

Personally, I don't think either system copes well at the extremes - the ideal solution is higher pixel count and and AA filter for both systems in my view.

But the one thing that the comparison does show is that the Foveon sensor is very capable of producing detailed images; more so than some people (who haven't used the camera) give it credit for.

The lesson? Experiment and evidence trump theory and posturing all of the time.

 




 
Quote
LOL! Well, there you are then. I could have sworn the bottom image, which shows obvious aliasing, was from the SD9. Considering the SD9 is 3.4mp, it would not be difficult to convince me, had the SD14 been used in the comparison, that the Foveon image would have been preferred to the 8mp Bayer type in all respects... at least in these examples. One would need to compare many different types of images to be certain about this though.

The fact that the 350D image appears to have greater accutance (like comparing a prime with a zoom) is probably attributable to its additional 2mp. The SD9 was always considered to be on a par with a 6mp Bayer type.
[a href=\"index.php?act=findpost&pid=150192\"][{POST_SNAPBACK}][/a]
Logged
My website and photo galleries: http://w
Pages: 1 ... 5 6 [7] 8 9 ... 11   Go Up