Luminous Landscape Forum
Site & Board Matters => About This Site => Topic started by: Ray on March 09, 2010, 10:02:52 pm
-
It's always seemed a bit absurd to me the way battle lines are drawn between MFDB and 35mm format when any discussion about comparative image quality is started.
It should be apparent to everyone who's been interested in photography for any significant period of time, that there's a strong connection between sensor size and image quality. Generally, the bigger the sensor, the better the image quality, at least in some major respects in not in all respects.
No-one could sensibly argue that a P&S camera produces image quality equal to that from an APS-C format. No-one could sensibly argue that an APS-C format produces image quality on a par with full frame 35mm, and likewise, no-one can sensibly argue that full frame 35mm produces image quality on a par with a DB which has two or more times the sensor area of 35mm.
If a sensor collects double the amount of light because it has double the area of another sensor, one would expect approximately a 1 stop increase in DR. If it has 4x the area, one could expect a 2 stop increase in DR, 8x the area 3 stops, 16x the area 4 stops and so on.
That progression of course assumes that all other factors contributing to DR are on an equal technological footing, and they rarely are. CMOS sensors are different to CCD sensors. My impression is that the photon-collecting diode on the CMOS sensor is smaller than the equivalent diode on a CCD sensor of equal pixel pitch.
That fact alone might explain why a DB, with double the sensor area of 35mm format, might have more than a one stop DR advantage. If each photon-collecting photodiode on the CCD pixel (of equal pitch) is double the area of the equivalent photodiode on a CMOS pixel, one might expect a 2 stop DR advantage, at base ISO.
However, the reason the photodiode on the CMOS pixel is smaller is in order to accommodate other processing devices on the sensor which improve signal-to-noise before the signal is digitised. I would gess that this is why, at the pixel level, a D3X has substantially higher DR than a P65+.
The D3X pixel is exactly the same size as that of the P65+. It's a testament to the technological prowess of Nikon and Sony that they've succeeded in gaining greater DR from a photodiode which is probably smaller than the photodiode on a much more expensive DB of equal pixel pitch.
However, the P65+ has many more pixels than the D3X. Comparing images of equal size, the increased DR of the D3X is marginal. Only 2/3rds of a stop at an 8"x12' print size.
But there's a question here which I've never seen addressed. When downsizing both images to an 8x12' size, as DXO does, one is throwing away image information from both cameras, in the example of the D3X and P65+, but one is throwing away more information from the P65+ image.
If one were to interpolate the D3X image to the same file size as the P65+ and then compare DR, what would be the result?
If one defines DR as the amount of meaningful image information in the deep shadows of an ETTR exposure, then it stands to reason that the P65+, with its significantly higher pixel count, might have a higher DR than the interpolated D3X image. At a guess, instead of 2/3rds of a stop lower DR (than the D3X) it might have 2/3rds of a stop higher DR. What do you think?
-
Ray
I don't think there were battle lines drawn - at least by those questioning the DR statements. However, there were some pretty extra-ordinary statements made by Mark and Michael - two very well respected photographers. As a consequence they were asked for evidence, which Michael suggests he and Mark are working on.
With regard to your discussion on bigger sensors being better than smaller ones - that may well be true for overall image quality, but my understanding is it has nothing to do with dynamic range, which is a function of the individual sensor "pixels" - and is generally a function of the capture area and the on-chip processing of the individual sensors.
I own a D700 right now, and in an informal test shooting inside and outside through a window, I was able to hold detail through between 8-9 stops. I think the D700 is a little better than average for DSLRS, so Michael and Mark's 6-7 stops of "usable DR" for DSLRs is probably not far off the mark (sorry Mark, that just flowed off my keyboard). The question is whether the MF can get 13 stops usable DR - something I seriously doubt until I see it.
In the end we all benefit from these debates and the internet keeps everyone honest.
Cheers
Rory
-
Hi,
I'm not really sure about this. Photographers see differences and those differences may be hard to explain. It may be that sometimes the talk is about microcontrast sometimes about tonality and now DR.
It is a bit unfortunate that DR is an engineering having an exact definition. So we start to talk about a usable DR, how do we define that.
The cameras are differently used.
- Mark has his P65 on an Alpa using a laser rangefinder quite often.
- MFDBs are used at low ISO - DSLRs are often used at relatively high ISO
- Personally I think lenses matter. DSLR lenses use to be complex designs. No one buys a 28/2.8 lens for a DSLR (well except Lloyd Chambers ;-) large aperture lenses contain a lot of extra glass. MFDBs used to have more conservative design and may be better finished.
My guess is that the DxO tests describe the sensors pretty well.
Comparing DSLRs and MFDBs is not that easy. Taking two different systems to the field is not very practical. For controlled conditions you would probably need a studio setup. With a well controlled studio setup you probably get into similar conditions DxO is using. Now, DxO is measuring on "raw-data". Photographers would typically use raw-converters and that induces a difference.
Images from Phase One backs contain a lot (around one megabyte?) of proprietary calibration information for each individual back, as explained in one of the LLVJ DVDs (I try to find out which). The phase converter can use this information to it's advantage.
Best regards
Erik
It's always seemed a bit absurd to me the way battle lines are drawn between MFDB and 35mm format when any discussion about comparative image quality is started.
It should be apparent to everyone who's been interested in photography for any significant period of time, that there's a strong connection between sensor size and image quality. Generally, the bigger the sensor, the better the image quality, at least in some major respects in not in all respects.
No-one could sensibly argue that a P&S camera produces image quality equal to that from an APS-C format. No-one could sensibly argue that an APS-C format produces image quality on a par with full frame 35mm, and likewise, no-one can sensibly argue that full frame 35mm produces image quality on a par with a DB which has two or more times the sensor area of 35mm.
If a sensor collects double the amount of light because it has double the area of another sensor, one would expect approximately a 1 stop increase in DR. If it has 4x the area, one could expect a 2 stop increase in DR, 8x the area 3 stops, 16x the area 4 stops and so on.
That progression of course assumes that all other factors contributing to DR are on an equal technological footing, and they rarely are. CMOS sensors are different to CCD sensors. My impression is that the photon-collecting diode on the CMOS sensor is smaller than the equivalent diode on a CCD sensor of equal pixel pitch.
That fact alone might explain why a DB, with double the sensor area of 35mm format, might have more than a one stop DR advantage. If each photon-collecting photodiode on the CCD pixel (of equal pitch) is double the area of the equivalent photodiode on a CMOS pixel, one might expect a 2 stop DR advantage, at base ISO.
However, the reason the photodiode on the CMOS pixel is smaller is in order to accommodate other processing devices on the sensor which improve signal-to-noise before the signal is digitised. I would gess that this is why, at the pixel level, a D3X has substantially higher DR than a P65+.
The D3X pixel is exactly the same size as that of the P65+. It's a testament to the technological prowess of Nikon and Sony that they've succeeded in gaining greater DR from a photodiode which is probably smaller than the photodiode on a much more expensive DB of equal pixel pitch.
However, the P65+ has many more pixels than the D3X. Comparing images of equal size, the increased DR of the D3X is marginal. Only 2/3rds of a stop at an 8"x12' print size.
But there's a question here which I've never seen addressed. When downsizing both images to an 8x12' size, as DXO does, one is throwing away image information from both cameras, in the example of the D3X and P65+, but one is throwing away more information from the P65+ image.
If one were to interpolate the D3X image to the same file size as the P65+ and then compare DR, what would be the result?
If one defines DR as the amount of meaningful image information in the deep shadows of an ETTR exposure, then it stands to reason that the P65+, with its significantly higher pixel count, might have a higher DR than the interpolated D3X image. At a guess, instead of 2/3rds of a stop lower DR (than the D3X) it might have 2/3rds of a stop higher DR. What do you think?
-
No one buys a 28/2.8 lens for a DSLR
you are very wrong... you probably need for example to check dslr forums where people sell/buy used glass... you will be very suprised, unless that suggestion of yours was really a joke akin to "6 stops of DR"
-
- Personally I think lenses matter. DSLR lenses use to be complex designs. No one buys a 28/2.8 lens for a DSLR (well except Lloyd Chambers ;-) large aperture lenses contain a lot of extra glass. MFDBs used to have more conservative design and may be better finished.
I don't think you can generalize like this. Assuming the comparison is between best-of-breed lenses for each system, I find it highly unlikely there's a significant difference in the "dynamic range" of the lenses. Today's best lenses have aspherical and low-dispersion elements, as well as extremely advanced coatings to reduce flare and ghosting.
There are complex lens designs in both formats. While a particular lens might have flare issues when shooting straight into the sun, more often than not this can be avoided, in which case I don't think the lens is going to be a limiting factor in dynamic range.
As for what DSLR users do or do not use, I think you'll find that many folks trying to get the most out of their high-res DSLR's are using primes. I know I shoot most of my landscapes with a 5-prime kit, of which only one of the lenses is faster than f/2.8 (and I'm quite confident the Zeiss 100 Makro-Planar has plenty of contrast despite its f/2 aperture).
-
In case anyone hasn't seen it, the following was added to the review yesterday...
Paragraph Removed – Editor
The struck-out paragraph below was part of the original article. It caused quite a bit of controversy on this site's forum and elsewhere. I initially added a comment to the bottom of the article, but that wasn't sufficient. Too many people's favourite ox had already been gored.
Please consider the paragraph below to be removed. There's no point in actually removing it, because once on the net, things last forever, and undoubtedly some zealots would consider it a coverup if we did so.
The reason that Mark and I have decided to remove it is not because we don't agree with its basic sentiment, but because it is serving as a distraction for the main point of the review, which is a comparison between two different medium format backs.
There is wide agreement among photographers that use or have used both formats that MF has an advantage of several F/stops in DR versus the small cameras. The precise number will vary depending on the specific camera and back being compared as well as the comparison methodology, but the difference is quite noticeable in actual images. From 30 feet away? Maybe not. Let's just chalk that phrase up as a bit of editorial hyperbole rather than something intended to be taken literally.
The issue of the differences between medium format and 35mm is a fascinating one though, and Mark and I intend on pursuing it in greater depth in the days ahead, and with more rigour than with a throw-away line or two.
-
Just wandering a little, very interesting to see Pentax back in the medium format business. 40mp 645D.
Price wise looks very competitive to rivals. I'm sure we'll see more on this one in due time.
-
You had to remove that comment, since it was at least contradictory with the article comparing the printout of a Canon G10 (not a 1Ds Mark-whatever) and a MF, where you stated that 40% of the people were not able to tell correctly which print was from which camera....
http://www.luminous-landscape.com/reviews/kidding.shtml (http://www.luminous-landscape.com/reviews/kidding.shtml)
-
Hi,
Sorry for that.
Let's put it this way. It seems hard to come up with any explanation that MFDBs would have more than one stop of advantage in DR. This may even be less, because DSLRs are using micro lenses. Photographers still see differences interpreted as DR. Not all light passing trough the lens is actually building image some part is scattered and shows up as flare. Figures I have seen may be around one percent. Now, one percent of flare would limit DR to about seven stops. Would flare amount to 0.1% about 10 stops of DR would be possible. Now, this percentage is of course dependent on how flare is calculated. It is a percentage of what?!
MFDB lenses used to be quite conservative designs. Normal lenses at around f/2.8 can be built with a small number of elements, a typical double Gauss normal lens usually has 6 elements in five groups. The lens elements are normally pretty small and fixed, they don't move around. Older lenses used to have metal tubing, well blackened and with internal baffles to keep the light from hitting the walls of the internal tubing. Modern lenses have more elements and may of them may moving around for zoom or internal focus. The elements may also be larger or more curved which may make them harder to fit inside the tubing.
Aspheric lenses don't reduce flare, but can improve on spherical aberration and fine detail contrast.
The Carl Zeiss 100/2 Macro Planar is one the finest lenses. Bernard Languillier, a frequent poster on this forum is using that lens. But that's certainly not the lens most people would use. Most people would use a Canon 24-70/2.8, a Canon 24/1.4L or the corresponding Nikon lenses. I use a Sony 24-70/2.8, mostly. I also have Sigma 12-24/4.5-5.6, Minolta 20/2.8, Minolta 50/1.4, Minolta 80-200/2.8 APO, Sony 70-300/4.5-5.6, Minolta 300/4 APO and 400/4.5 APO. I don't have a Zeiss 85/1.4 or a 135/1.8, sure they are great lenses. But I really want to have the flexibility of the zooms and fast autofocus.
Best regards
Erik
I don't think you can generalize like this. Assuming the comparison is between best-of-breed lenses for each system, I find it highly unlikely there's a significant difference in the "dynamic range" of the lenses. Today's best lenses have aspherical and low-dispersion elements, as well as extremely advanced coatings to reduce flare and ghosting.
There are complex lens designs in both formats. While a particular lens might have flare issues when shooting straight into the sun, more often than not this can be avoided, in which case I don't think the lens is going to be a limiting factor in dynamic range.
As for what DSLR users do or do not use, I think you'll find that many folks trying to get the most out of their high-res DSLR's are using primes. I know I shoot most of my landscapes with a 5-prime kit, of which only one of the lenses is faster than f/2.8 (and I'm quite confident the Zeiss 100 Makro-Planar has plenty of contrast despite its f/2 aperture).
-
Hi,
Sorry for that.
Let's put it this way. It seems hard to come up with any explanation that MFDBs would have more than one stop of advantage in DR. This may even be less, because DSLRs are using micro lenses. Photographers still see differences interpreted as DR. Not all light passing trough the lens is actually building image some part is scattered and shows up as flare. Figures I have seen may be around one percent. Now, one percent of flare would limit DR to about seven stops. Would flare amount to 0.1% about 10 stops of DR would be possible. Now, this percentage is of course dependent on how flare is calculated. It is a percentage of what?!
MFDB lenses used to be quite conservative designs. Normal lenses at around f/2.8 can be built with a small number of elements, a typical double Gauss norma lens usually has 6 elements in five groups. The lens elements are normally pretty small and fixed, they don't move around. Older lenses used to have metal tubing, well blackened and with internal baffles to keep the light from hitting the walls of the internal tubing. Modern lenses have more elements and may of them may moving around for zoom or internal focus. The elements may also be larger or more curved which may make them harder to fit inside the tubing.
Aspheric lenses don't reduce flare, but can improve on spherical aberration and fine detail contrast.
The Carl Zeiss 100/2 Macro Planar is one the finest lenses. Bernard Languillier, a frequent poster on this forum is using that lens. But that's certainly not the lens most people would use. Most people would use a Canon 24-70/2.8, a Canon 24/1.4L or the corresponding Nikon lenses. I use a Sony 24-70/2.8, mostly. I also have Sigma 12-24/4.5-5.6, Minolta 20/2.8, Minolta 50/1.4, Minolta 80-200/2.8 APO, Sony 70-300/4.5-5.6, Minolta 300/4 APO and 400/4.5 APO. I don't have a Zeiss 85/1.4 or a 135/1.8, sure they are great lenses. But I really want to have the flexibility of the zooms and fast autofocus.
Best regards
Erik
You have touched on an often overlooked and critical point about DR. DR is often lens, not sensor, limited. Like you say, it's all about internal reflections and flare. Simple prime lenses with few elements/groups and good internal baffling generally allow more DR than a complicated zoom with a UV filter on it.
-
Hi,
I made some similar experiments with APS-C and full format 135 DSLRs. Sometimes the difference may be hard to see in prints. Depends much on detail size.
I touch on the issue in this article: http://echophoto.dnsalias.net/ekr/index.ph...xels-do-we-need (http://echophoto.dnsalias.net/ekr/index.php/photoarticles/24-how-many-megapixels-do-we-need)
This page has some interesting links: http://echophoto.dnsalias.net/ekr/index.ph...vs-mfdb-vs-film (http://echophoto.dnsalias.net/ekr/index.php/photoarticles/25-dslr-vs-mfdb-vs-film)
Best regards
Erik
You had to remove that comment, since it was at least contradictory with the article comparing the printout of a Canon G10 (not a 1Ds Mark-whatever) and a MF, where you stated that 40% of the people were not able to tell correctly which print was from which camera....
http://www.luminous-landscape.com/reviews/kidding.shtml (http://www.luminous-landscape.com/reviews/kidding.shtml)
-
Michael and Mark say they are going to do more work and come back to this topic, and I hope they do, and get it done before the new round of Canon and Nikon and the new Pentax MF high-res cameras are released...
I have a problem with "real world" discussions, however. I'm not a scientist, but I have studied the philosophy of science, and would have interesting and detailed discussions about scientific method with my late wife, who had a PhD in immunology, which is an intricate and complicated and often confusing field of study. Ultimately the problem with "real world" views is that they confuse issues. It's quite possible that Mark's anecdote about seeing differences in a print at thirty feet is entirely accurate -- in that people can see differences - but those differences don't really involve DR, even though people sincerely believe they do. That simply has to do with the way the brain works -- a much large sensor delivers, say, terrific microcontrast and subtle color across a wide dynamic range, and people may assume that the dynamic range is wider that it actually is, because they are being influenced by other factors. Basic color theory texts routinely demonstrate "perceived" differences in DR by placing different colors adjacent to a test color...and the test color's brightest or dullness seems to fluctuate depending upon the adjacent color, although that color does not change at all. Separating out these exterior influences can be nearly impossible in "real world" views.
The other thing that I've learned as a regular reader of Nature, an addition to my relationship with a serious scientist, is that test results and measurements often come out the way the testers wish they would...so it's perfectly possible that serious engineers for MF companies believe their dynamic range is 6-7 stops wider than 35mm-equiv digital, because they *wish* that were true. In my experience, the best results come from sources who don't care about the outcome of the tests...one reason, for example, that I've been more than a bit wary about the Tour de France drug-testing methods involving people like Floyd Landis. I would have been much more confident of the outcome if the tests were done, say, in Japan.
So, I hope if this is further explored, that the testing methods will be as rigorous as possible.
JC
-
Michael and Mark say they are going to do more work and come back to this topic, and I hope they do, and get it done before the new round of Canon and Nikon and the new Pentax MF high-res cameras are released...
I have a problem with "real world" discussions, however. I'm not a scientist, but I have studied the philosophy of science, and would have interesting and detailed discussions about scientific method with my late wife, who had a PhD in immunology, which is an intricate and complicated and often confusing field of study. Ultimately the problem with "real world" views is that they confuse issues. It's quite possible that Mark's anecdote about seeing differences in a print at thirty feet is entirely accurate -- in that people can see differences - but those differences don't really involve DR, even though people sincerely believe they do. That simply has to do with the way the brain works -- a much large sensor delivers, say, terrific microcontrast and subtle color across a wide dynamic range, and people may assume that the dynamic range is wider that it actually is, because they are being influenced by other factors. Basic color theory texts routinely demonstrate "perceived" differences in DR by placing different colors adjacent to a test color...and the test color's brightest or dullness seems to fluctuate depending upon the adjacent color, although that color does not change at all. Separating out these exterior influences can be nearly impossible in "real world" views.
The other thing that I've learned as a regular reader of Nature, an addition to my relationship with a serious scientist, is that test results and measurements often come out the way the testers wish they would...so it's perfectly possible that serious engineers for MF companies believe their dynamic range is 6-7 stops wider than 35mm-equiv digital, because they *wish* that were true. In my experience, the best results come from sources who don't care about the outcome of the tests...one reason, for example, that I've been more than a bit wary about the Tour de France drug-testing methods involving people like Floyd Landis. I would have been much more confident of the outcome if the tests were done, say, in Japan.
So, I hope if this is further explored, that the testing methods will be as rigorous as possible.
JC
Thank you John,
That is indeed an interesting point of view. I read your post with attention and I think you brought a very interesting focus on the matter.
I tend to agree in many of the points you made.
Cheers,
Fred.
-
Microlenses increase QE and are one of the main factors behind the better high ISO peformance of DSLRs. DR is unaffected: the well capacity is unchanged, the well just fills faster, and the read noise remains the same.
Hardware (on chip) binning is a sure method to increase DR. That's what "real" stuff uses when maximizing DR is required (http://www.marresearch.com/products.mx-series.html). Software binning can improve DR somewhat as well (see bottom of this page http://www.theta-system.com/sis.html) (http://www.theta-system.com/sis.html)) but that depends on a few other factors. Since it seems polite to make a difference between engineering QE, DR and photographic "high iso" sensitivity and "photographic" DR, "photographic binning" can play a role as well. If you shoot aunt Eunice with a DSLR and a MFDB and print the portraits at the same size or compare them on the same screen, you've "binned" the MFDB sensor relatively to the DSLR and probably gained a bit of DR/transition smoothness in the process.
Anyway, one of the best analysis of the issues involved is Emil's work, which was posted here earlier.
http://theory.uchicago.edu/~ejm/pix/20d/te...oise/index.html (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html)
Of course, lenses do play a very significant role: the sensor doesn't sample Aunt Eunice, it samples whatever image the lens provides. Engineering DR remains unaffected, but our perception of the rendered DR versus the actual DR of the scene is what matters most...
PS: I couldn't help thinking that someone who glanced quickly at Emil's article and charts might have had the impression that DSLRs DR was in the 7-8 stops range. ;-)
-
With regard to your discussion on bigger sensors being better than smaller ones - that may well be true for overall image quality, but my understanding is it has nothing to do with dynamic range, which is a function of the individual sensor "pixels" - and is generally a function of the capture area and the on-chip processing of the individual sensors.
Rory, it seems clear that DR is a function of both individual pixels and over all sensor size. Sorry to keep referring to DXO test results, but there's an interesting and relevant comparison between the 8mp Canon 20D and the recent 5D2 at:
http://www.dxomark.com/index.php/eng/Image.../(brand2)/Canon (http://www.dxomark.com/index.php/eng/Image-Quality-Database/Compare-cameras/(appareil1)/178%7C0/(appareil2)/305%7C0/(onglet)/0/(brand)/Canon/(brand2)/Canon)
Just as the pixel size of the P65+ is the same as that of the D3X, the pixel size of the 5D2 is the same as that of the 20D, except the 5D2 and 20D employ very similar CMOS technology so the relationship between DR and sensor size is less likely to be affected by other design factors.
What I find interesting is just how close the DR of the 20D pixel is to that of the 5D2 pixel despite the intervening years of technological development. It's almost as though Canon took the old 20D sensor and just multiplied its area by 2.6 to get the full frame 5D2.
The DR of the 5D2 pixel is very marginally better, by about 1/4th of a stop all the way up to ISO 3200 which is the highest ISO setting on the 20D. The slight improvement is possibly due to a further narrowing of the gap between microlenses and perhaps a slight increase in the size of the photon-collecting diode in the 5D2 pixel. However, the tonal range is exactly the same, and very susrprisingly the color sensitivity of the 20D pixel is actually slightly better than that of the 5D2.
Now DXO Mark also provide results at a normalised print size of 8x12" at 300ppi, which so happens to be the exact print size from a 20D at 100%. At that print size, the performance of the 5D2 takes a leap. All the parameters of SNR, DR, tonal range, color sensitivity are significantly better than the 20D. DR is close to one stop better, for example.
One might reasonably deduce that a P65+ would have close to a 1 stop DR advantage over the D3X when the P65+ file is downsized to the same size as the D3X file provided that the technologies employed in both sensors are the same or very similar, which of course they are not.
-
What I find interesting is just how close the DR of the 20D pixel is to that of the 5D2 pixel despite the intervening years of technological development. It's almost as though Canon took the old 20D sensor and just multiplied its area by 2.6 to get the full frame 5D2.
The DR of the 5D2 pixel is very marginally better, by about 1/4th of a stop all the way up to ISO 3200 which is the highest ISO setting on the 20D. The slight improvement is possibly due to a further narrowing of the gap between microlenses and perhaps a slight increase in the size of the photon-collecting diode in the 5D2 pixel. However, the tonal range is exactly the same, and very susrprisingly the color sensitivity of the 20D pixel is actually slightly better than that of the 5D2.
Now DXO Mark also provide results at a normalised print size of 8x12" at 300ppi, which so happens to be the exact print size from a 20D at 100%. At that print size, the performance of the 5D2 takes a leap. All the parameters of SNR, DR, tonal range, color sensitivity are significantly better than the 20D. DR is close to one stop better, for example.
I see your point Ray. I presume the interpolation of the larger files reduce noise and therefore increase DR.
-
The other thing that I've learned as a regular reader of Nature, an addition to my relationship with a serious scientist, is that test results and measurements often come out the way the testers wish they would...so it's perfectly possible that serious engineers for MF companies believe their dynamic range is 6-7 stops wider than 35mm-equiv digital, because they *wish* that were true. In my experience, the best results come from sources who don't care about the outcome of the tests...one reason, for example, that I've been more than a bit wary about the Tour de France drug-testing methods involving people like Floyd Landis. I would have been much more confident of the outcome if the tests were done, say, in Japan.
So, I hope if this is further explored, that the testing methods will be as rigorous as possible.
JC
That's a good point, John, and the reason for the necessity of double-blind testing and the repeatability of testing and experimentation by other parties who may not have the same vested interests and agenda.
There's also the complication of choice of RAW converter, the sorts of processing adjustments used and the fact that one particular brand of raw converter may favour one particular brand of camera. Remember RSP (RawShooter premium) that was taken over by Adobe? It used to be my preferred RAW converter for certain landscapes, especially from the 5D.
I preferred it because it gave me results with just a few adjustments of the sliders, which ACR couldn't provide with the same ease. I once tried to emulate an RSP conversion by making as many adjustments in ACR to get the RAW image looking as close as possible, and then making further adjustments to the converted image in Photoshop. I eventually got the ACR conversion looking very similar to the RSP conversion, but it was a lot of work.
I wouldn't expect owners of expensive MFDB equipment to work too hard to get an image from a much less expensive 35mm camera looking almost as good.
-
I feel that one key aspect in the perception of DR is the way the raw converters handles local contrast.
We know for a fact that both MFDB and high end DSLR have too much DR for a pleasing rendering without some tweaking. Images look flat when you attempts to map say 11 stops of DR on an 8 bit display.
So you need to work with local contract to generate pleasing results.
We know that the highlight/shadow sliders ot leading raw converters do play with local contrast, to the extend that they sometime generate halos similar to what one gets when using Photomatix.
How much of this is automated when you open a P65+ file in C1 is a topic on which I would be interested in getting more information.
Cheers,
Bernard
-
The other thing that I've learned as a regular reader of Nature, an addition to my relationship with a serious scientist, is that test results and measurements often come out the way the testers wish they would...so it's perfectly possible that serious engineers for MF companies believe their dynamic range is 6-7 stops wider than 35mm-equiv digital, because they *wish* that were true. In my experience, the best results come from sources who don't care about the outcome of the tests...one reason, for example, that I've been more than a bit wary about the Tour de France drug-testing methods involving people like Floyd Landis. I would have been much more confident of the outcome if the tests were done, say, in Japan.
So, I hope if this is further explored, that the testing methods will be as rigorous as possible.
This is a very good point. Science is full of examples of how even large groups of honest and good scientist can fool them self.
The book "The Undergrowth of Science - delusion, self-deception and human frailty" by Walter Gratzer gives many examples of how scientists, and even nations, have failed in their pursue of science.
Ronny
-
The other thing that I've learned as a regular reader of Nature, an addition to my relationship with a serious scientist, is that test results and measurements often come out the way the testers wish they would...so it's perfectly possible that serious engineers for MF companies believe their dynamic range is 6-7 stops wider than 35mm-equiv digital, because they *wish* that were true. In my experience, the best results come from sources who don't care about the outcome of the tests...one reason, for example, that I've been more than a bit wary about the Tour de France drug-testing methods involving people like Floyd Landis. I would have been much more confident of the outcome if the tests were done, say, in Japan.
So, I hope if this is further explored, that the testing methods will be as rigorous as possible.
JC
My background is in biochemistry and I've done enough lab work over the years and in my current professional career to understand what is good science and what is not. My read of the DxO data is that it's an engineering approach and there is very little in terms of subjective endpoints to the research. this is the real key to looking at any research. Clearly access to data sets and methodology is important for independent corroboration (this the role of the US Food and Drug Administration in reviewing new pharmaceuticals for approval).
All this being said, there is some work that we all can do. If you go back to the closed thread, there were some nice posts about possible experiments that can be done (for example, see the Image66 post at: http://luminous-landscape.com/forum/index....2074&st=160 (http://luminous-landscape.com/forum/index.php?showtopic=42074&st=160) ). One can take any of the popular printer evaluation images and do this type of experiment to see what the useful photographic range of a camera might be. The classic B&W zone exposure test can also be done using a textured mono-color target and expose throughout the range, though this would of course involve a B&W conversion for evaluation. One could test a prime lens vs a zoom lens to see if there is a major difference as some have postulated. We could also build a mini data set for various cameras as well. Providing the weekend doesn't get rained out here in the mid-Atlantic, I'm going to try both approaches using natural light (I don't have any kind of interior set up that I could use) with my Nikon D300 (and maybe even Canon S90) to see what the ranges are. Will post protocols and results on my website and make RAW files available as well.
Alan
-
Dynamic range is complex because it is so dependent upon the detection situation. For example, at high light throughput levels lens flare becomes limiting. At low levels, the detector read noise is limiting for short exposures and shot noise for long exposures. This all interacts with wavelength, the angle at which rays strike the detector, detector surface treatments....... Given all this, just how are we to set up the situation to yield a meaningful (replicable and consistent) DR value?
Engineers use a definition that removes uncontrolled variables as much as possible. It allows them to compare detectors and entire optical trains with some degree of objectivity. Note I say only some degree, because at high bit densities we are looking at very minute differences that tend to fall within the noise floor of the measurement technology. There will be some uncertainty.
Point is, don't expect an engineering DR figure to have much to do with your photography. The DR number reflects rigorously defined measurement protocols and specifically excludes subjective factors such as perceived image quality.
The DR figure is very useful with well defined applications. For example, if you needed an imaging system for low light you would know to select a detector with high QE, slow readout for read noise and cooling for shot noise. Chip packages are well specified on these parameters so you could use DR values to make a direct comparison. Similarly, the lens should have low internal fluorescence/reflectance and various other suitable characterisitcs. Again, these data are available and scientists select optical systems on that basis every day.
The problem is that we do not have a clear definition of what we need for photography - nor do we have an accepted measurement protocol for DR. Therefore, it is not entirely reasonable to expect to be able to look up a CCD or CMOS data set that correlates well with perceived image qualtiy. We are left to make subjective decisions and tend to depend on reviews from trusted sources. We are, therefore, doomed to the endless arguments.
-
Science is full of examples of how even large groups of honest and good scientist can fool them self.
Ronny
Correct interpretation of data has always been a problem for many working scientists who are not properly trained in statistical reasoning and pattern recognition. I shall give an example, with hopefully some close to realistic numbers. Suppose that the prevalence rate of HIV among a population is 0.01%, an HIV test procedure is 99.8% sensitive (i.e., 99.8% times it is right on those people whom we know have been inflicted with HIV), and the false positive rate is 0.01% (i.e., those which are identified to have HIV where as in actuality they didn't have it.). Then the probability that a person really has HIV if the test says that the person has HIV is actually almost 50%. I.e., you can flip a coin and decide if that person has HIV or not! However, don't get scared, because, the probability that a person does not have HIV if the test indicates no HIV is 99.999%, i.e., you are almost certain that that person does not have HIV.
The tragic incident of Sally Clark (http://en.wikipedia.org/wiki/Sally_Clark) in England is an example when she was thought of murdering her children which the Royal Statistical Society identified as a misuse of statistics in court procedures.
Many of the signal processing algorithms are special cases of Bayesian statistics. For e.g., our beloved Lucy-Richardson approach for image sharpening. However, many signal processing algorithms make simplifying assumptions in many procedures that make the interpretation of data different.
-
Correct interpretation of data has always been a problem for many working scientists who are not properly trained in statistical reasoning and pattern recognition. I shall give an example, with hopefully some close to realistic numbers. Suppose that the prevalence rate of HIV among a population is 0.01%, an HIV test procedure is 99.8% sensitive (i.e., 99.8% times it is right on those people whom we know have been inflicted with HIV), and the false positive rate is 0.01% (i.e., those which are identified to have HIV where as in actuality they didn't have it.). Then the probability that a person really has HIV if the test says that the person has HIV is actually almost 50%. I.e., you can flip a coin and decide if that person has HIV or not! However, don't get scared, because, the probability that a person does not have HIV if the test indicates no HIV is 99.999%, i.e., you are almost certain that that person does not have HIV.
That is a good demonstration of Bayesian analysis. In practice, clinicians screen for HIV with an ELISA test, which is sensitive but not that specific. If the ELISA is positive, the diagnosis is confirmed with a more specific Western Blot test.
The tragic incident of Sally Clark (http://en.wikipedia.org/wiki/Sally_Clark) in England is an example when she was thought of murdering her children which the Royal Statistical Society identified as a misuse of statistics in court procedures.
Many of the signal processing algorithms are special cases of Bayesian statistics. For e.g., our beloved Lucy-Richardson approach for image sharpening. However, many signal processing algorithms make simplifying assumptions in many procedures that make the interpretation of data different.
An excellent example of the misuse of statistics. One of my professors used to warn us, "Figures do not lie, but liars can figure". Fortunately, the death penalty was not imposed on Ms. Clark. Human perception is not that reliable, either in criminology or in evaluation of photographic images. In both fields, objective measurements are needed. In the USA we have numerous examples of persons on death row for murder after a positive eye witness identification, but who have been proven innocent by DNA analysis. Although the consequences of faulty perception are not as grave when one is evaluating dynamic range of photographs as with the identification of murder suspects, an objective scientific measurement is desirable.
-
Dynamic range is complex because it is so dependent upon the detection situation. For example, at high light throughput levels lens flare becomes limiting. At low levels, the detector read noise is limiting for short exposures and shot noise for long exposures. This all interacts with wavelength, the angle at which rays strike the detector, detector surface treatments....... Given all this, just how are we to set up the situation to yield a meaningful (replicable and consistent) DR value?
Engineers use a definition that removes uncontrolled variables as much as possible. It allows them to compare detectors and entire optical trains with some degree of objectivity. Note I say only some degree, because at high bit densities we are looking at very minute differences that tend to fall within the noise floor of the measurement technology. There will be some uncertainty.
Point is, don't expect an engineering DR figure to have much to do with your photography. The DR number reflects rigorously defined measurement protocols and specifically excludes subjective factors such as perceived image quality.
The DR figure is very useful with well defined applications. For example, if you needed an imaging system for low light you would know to select a detector with high QE, slow readout for read noise and cooling for shot noise. Chip packages are well specified on these parameters so you could use DR values to make a direct comparison. Similarly, the lens should have low internal fluorescence/reflectance and various other suitable characterisitcs. Again, these data are available and scientists select optical systems on that basis every day.
The problem is that we do not have a clear definition of what we need for photography - nor do we have an accepted measurement protocol for DR. Therefore, it is not entirely reasonable to expect to be able to look up a CCD or CMOS data set that correlates well with perceived image qualtiy. We are left to make subjective decisions and tend to depend on reviews from trusted sources. We are, therefore, doomed to the endless arguments.
Speaking personally, I've never had a problem in determining subjectively whether one camera produces a higher DR than another. It was always apparent to me that images from negative film had a higher DR than images from slide film. Likewise, it was very apparent that my first P&S camera (the Sony T1) had worse DR than my Canon D60 DSLR, and that my second DSLR (the 20D) had much better DR than my D60 above base ISO, but not much difference at base ISO.
It was also apparent that my first full frame DSLR, the 5D, did not have the expected increase in DR compared with the 20D. In fact it seemed worse. The deepest shadows displayed ugly banding. I returned the unit for a replacement which I considered better but still not entirely satisfactory. The chief advantages of the 5D were the flow-on effects of the larger sensor with substantially greater pixel count.
At same image or print size as the 20D there seemed to be better color, lower noise, better tonal range and of course higher resolution at big print sizes and as a consequence better DR at such print sizes.
Your point about the engineering specification for DR having little to do with the perception of DR in the photograph seems only partly true to me. The question that should be asked is; is there any reason why such so-called engineering specifications (as in DXO figures) are not valid for the purpose of comparison? I mean, we're not talking about the sensor manufacturer's engineering specification for the sensor itself, unattached to a camera.
The remarkable thing about DXO results for the D3X is that it is claimed the D3X has 1 & 1/3rd stops higher DR than the P65+ at the pixel level. The D3X pixel is exactly the same size as the P65+ pixel, yet Nikon have employed such advanced technology that their pixel, despite its probably having the disadvantage of a smaller photon-collecting diode and therefore able to collect less light, actually has a higher DR.
The actual figure that DXO specify, 12.84 EV, might be unrealistically high from the perspective of the photographer and the viewer. The image detail and quality in that 13th stop might be be just awful and totally useless, and would therefore normally be clipped to black during processing, except in artistic shots like this which attempts to turn the ugly banding of the 5D into beauty; an accident when the flash did not fire.
[attachment=20834:Temple_b..._Ayudhya.jpg]
However, for the purposes of comparison, one would examine the degree of awfulness in that 13th stop. According to DXO, the P65+ image (of identical scene and lighting of course) would be even more awful in the 13th stop than the D3X, and no doubt more awful in the 12th and 11th stop.
The question then becomes, at what stop is the detail and quality useful so that it could be preserved in the print instead of being clipped to black? Perhaps in the 9th stop? Real world comparisons should examine such issues. The fact that the D3X pixel is the same size as the P65+ pixel makes such comparisons very easy. I'm really surprised no-one's taken the trouble to compare the D3X with the P65+, at the pixel level, to either confirm or refute the DXO claims.
Because the pixel size is the same, all one has to do is use the same focal length of lens on both cameras, shoot the same 'high SBR' scene with the same lighting, from the same position, and then crop the P65+ image to the same FOV as the D3X image (and same aspect ratio). Both images will then have the same file size and be comprisied of the same number of pixels. DR comparison would be easy, provided the exposures are correct with regard to ETTR. What could be easier! There even no need to adjust f stop for equal DoF, always a contentious issue.
Okay! Okay! Lens flare. I must confess that I didn't realise that lens flare could be such a limiting factor on DR. We've all experienced the annoying effects of lens flare when the camera angle is too close to the direct rays of the sun, but the fact that lens flare may reduce DR when the sun isn't even in sight, should be a factor taken into consideration when comparing the DR of different format cameras that use different lenses.
In the light of such information provided in the other current thread on this issue (Dynamice Range and DXO), it would seem that any thorough comparison between the D3X and P65+ should first examine the flare characteristics of the lenses used with both cameras.
I've long been an advocate of specific lens testing by the manufacturer (or contractor) of each lens sold, because we all know that lens quality variability amongst copies of the same model of lens is an issue. I would now add a further requirement for a 'flare test' of such individual lens copies.
-
Okay! Okay! Lens flare. I must confess that I didn't realise that lens flare could be such a limiting factor on DR. We've all experienced the annoying effects of lens flare when the camera angle is too close to the direct rays of the sun, but the fact that lens flare may reduce DR when the sun isn't even in sight, should be a factor taken into consideration when comparing the DR of different format cameras that use different lenses.
In the light of such information provided in the other current thread on this issue (Dynamice Range and DXO), it would seem that any thorough comparison between the D3X and P65+ should first examine the flare characteristics of the lenses used with both cameras.
I've long been an advocate of specific lens testing by the manufacturer (or contractor) of each lens sold, because we all know that lens quality variability amongst copies of the same model of lens is an issue. I would now add a further requirement for a 'flare test' of such individual lens copies.
I would reckon that lens flare might have a much bigger impact than you might think. I did some tests (on film) a couple of years ago using two apparently identical Zeiss 80mm Planars (both silver C lenses ca 1971), off a tripod on a static subject outdoors under diffused overcast conditions (no deep shadows). I was using the 'Blad Pro Hood, so no direct light fell on the lens. One lens was T* coated, the other had the older single-coating. When printed, there was no difference in definition, but a noticeable difference in DR in favour of the T* - about half a stop, I would guess, between highlight and shadows. So I don't really see how you can compare camera sensor DR without using exactly the same lens on both, which is usually impossible.
John
-
I believe that lens flare will actually contribute to an effective INCREASE in usable dynamic range of the scene.
A bright sunny day will give us extremes in exposures from highlights to deep shadows. If you expose for the highlights, the shadows will drop below the sensitivity curves of the sensor or film. But with a bit of lens flare, you will raise the brightness level of the shadows to fall into the acceptance range of the sensor or film. (lens flare being internal reflections not resulting in rings, ghosts and other artifacts)
In the darkroom, we use "pre-flashing" to keep highlights from blowing out in B&W prints, and pre-flashing is also a tried and true technique with large-format film to recover details in the shadows.
This is also why older single-coated lenses are usually considered to be superior for B&W photography than the later highly flare-resistant and contrasty newer lenses.
With digital imaging, we have a rather narrow range of lumenance values which can be recorded. There is no toe or shoulder to speak of, but some cameras, like my Olympus E-1 which I performed the tests with, have a distinct toe whereas my old Minolta A1 had a distinct shoulder which can give the illusion of increased dynamic range when in reality the "straight-line sections" of the response/capture curves are similar, if not identical. Because of this narrow range of values which can be recorded, it is our responsibility as photographers to modify the lighting or the incoming scene to the camera to restrict the values to that which can be recorded by the medium.
Lower-contrast lenses are a means to "pre-sensitize" the sensor/film to give the medium an extended toe. If you subscribe to the "expose to the right" exposure method, then your highlights don't change, but the lower-contrast lens will pull up the shadows into a range either above the noise-floor or at least to a point where posterization isn't a problem.
Just as an aside, but some cameras, such as the Olympus E-1 and the Kodak 14n induced random noise on the image data to provide smoother tonal and brightness transitions throughout the range of exposure values. Other cameras provide ultra-clean images but have a non-linear addition of noise in the low values. Unfortunately, this is electronic noise which is visible when boosting the shadows or the ISO. The Kodak/Olympus method means that you'll have noise even in the highest values, but it is consistent throughout the entire range--just like film.
From what I've been seeing, it looks to me like some of the MFDB camera systems also have induced noise on the image data to mask the nasties as well as providing superior micro-contrast.
From the beginning of digital time, I've maintained that CMOS images just didn't look quite right. There was something off which is difficult to explain or describe. As CMOS isn't typically a technology used in the MFDB systems, what we may be seeing is NOT the difference between dynamic range, pixel pitch or even formats, but we're seeing the difference between a CMOS imager and a CCD imager.
It is very difficult to know for sure, as few imager manufacturers will publish specifications on the chips. For all we know, the CMOS images are unable to capture some colors or brightness values and they are being derived. It's hard to say since these are guarded secrets from us photographers. There are obviously differences, though. When working in Lightroom or your converter of choice, we have to bend the brightness curves to get one camera to match another. This bending of the curves is the stealing bits (dynamic range and micro-contrast) from one part of the curve to reassign to another part of the curve. One thing is a guarantee--no two imager designs has the same response curves and the support electronics in the camera reassign the values to a more consistent form BEFORE the raw file is written. Raw isn't raw.
Ken Norton
www.zone-10.com
-
Hi,
I see your point, but I don't agree. The effect you see is probably for real. It's just that I prefer to have a high contrast flare resistant lens, a sensor that's capable of high DR and tame the "raw image" in development.
The sensor is essentially a linear device. Any toe and shoulder characteristics are essentially added in processing, either in camera processing for JPEG or "raw" processing after the fact.
Best regards
Erik
I believe that lens flare will actually contribute to an effective INCREASE in usable dynamic range of the scene.
A bright sunny day will give us extremes in exposures from highlights to deep shadows. If you expose for the highlights, the shadows will drop below the sensitivity curves of the sensor or film. But with a bit of lens flare, you will raise the brightness level of the shadows to fall into the acceptance range of the sensor or film. (lens flare being internal reflections not resulting in rings, ghosts and other artifacts)
In the darkroom, we use "pre-flashing" to keep highlights from blowing out in B&W prints, and pre-flashing is also a tried and true technique with large-format film to recover details in the shadows.
This is also why older single-coated lenses are usually considered to be superior for B&W photography than the later highly flare-resistant and contrasty newer lenses.
With digital imaging, we have a rather narrow range of lumenance values which can be recorded. There is no toe or shoulder to speak of, but some cameras, like my Olympus E-1 which I performed the tests with, have a distinct toe whereas my old Minolta A1 had a distinct shoulder which can give the illusion of increased dynamic range when in reality the "straight-line sections" of the response/capture curves are similar, if not identical. Because of this narrow range of values which can be recorded, it is our responsibility as photographers to modify the lighting or the incoming scene to the camera to restrict the values to that which can be recorded by the medium.
Lower-contrast lenses are a means to "pre-sensitize" the sensor/film to give the medium an extended toe. If you subscribe to the "expose to the right" exposure method, then your highlights don't change, but the lower-contrast lens will pull up the shadows into a range either above the noise-floor or at least to a point where posterization isn't a problem.
Just as an aside, but some cameras, such as the Olympus E-1 and the Kodak 14n induced random noise on the image data to provide smoother tonal and brightness transitions throughout the range of exposure values. Other cameras provide ultra-clean images but have a non-linear addition of noise in the low values. Unfortunately, this is electronic noise which is visible when boosting the shadows or the ISO. The Kodak/Olympus method means that you'll have noise even in the highest values, but it is consistent throughout the entire range--just like film.
From what I've been seeing, it looks to me like some of the MFDB camera systems also have induced noise on the image data to mask the nasties as well as providing superior micro-contrast.
From the beginning of digital time, I've maintained that CMOS images just didn't look quite right. There was something off which is difficult to explain or describe. As CMOS isn't typically a technology used in the MFDB systems, what we may be seeing is NOT the difference between dynamic range, pixel pitch or even formats, but we're seeing the difference between a CMOS imager and a CCD imager.
It is very difficult to know for sure, as few imager manufacturers will publish specifications on the chips. For all we know, the CMOS images are unable to capture some colors or brightness values and they are being derived. It's hard to say since these are guarded secrets from us photographers. There are obviously differences, though. When working in Lightroom or your converter of choice, we have to bend the brightness curves to get one camera to match another. This bending of the curves is the stealing bits (dynamic range and micro-contrast) from one part of the curve to reassign to another part of the curve. One thing is a guarantee--no two imager designs has the same response curves and the support electronics in the camera reassign the values to a more consistent form BEFORE the raw file is written. Raw isn't raw.
Ken Norton
www.zone-10.com
-
[quote name='image66' date='Mar 12 2010, 02:21 PM' post='352945']
I believe that lens flare will actually contribute to an effective INCREASE in usable dynamic range of the scene.
A bright sunny day will give us extremes in exposures from highlights to deep shadows. If you expose for the highlights, the shadows will drop below the sensitivity curves of the sensor or film. But with a bit of lens flare, you will raise the brightness level of the shadows to fall into the acceptance range of the sensor or film. (lens flare being internal reflections not resulting in rings, ghosts and other artifacts)
Ken, I think you provide an excellent example of why working photographers should not obsess about dynamic range. The flare effect that you describe is a compression of DR, not an expansion. Never mind. There are worse things going on in real life imaging. For example, pixels are not independent of each other. If you shine a bright light on a pixel to the left of center, the pixels at center will show a rise. This type of local blooming effect - which you could view as a degradation of local contrast - severely limits DR under some conditions. There are lots of problems in detector application, and the solutions vary according to what the end user wants. A portrait photgrapher and a microscopist have completely different needs. My point is that we can't allow such situational factors to affect what should be a basic performance measurement that allows us to compare devices. That comparison happens before the subjective part.
It is your eye which tells you a camera is doing what you want. In your case that appears to be a sort of integral gain riding and that could certainly expand exposure lattitude. My point is that, unless photographers can specify exactly what it is they want, they will have a great deal of trouble establishing camera usability from engineering measurements. Scientific imagers tend to specify. Creative photographers - not so much. Instead they end up comparing what they like as much as what different cameras are doing and there is nothing wrong with that. It is inherently meaningful - just difficult to quantify.
By the way, I do not think there are any secrets or conspiracies in the detector market. You could probably find the engineering specs for any extant chip package if you look in the right places.
-
I would reckon that lens flare might have a much bigger impact than you might think. I did some tests (on film) a couple of years ago using two apparently identical Zeiss 80mm Planars (both silver C lenses ca 1971), off a tripod on a static subject outdoors under diffused overcast conditions (no deep shadows). I was using the 'Blad Pro Hood, so no direct light fell on the lens. One lens was T* coated, the other had the older single-coating. When printed, there was no difference in definition, but a noticeable difference in DR in favour of the T* - about half a stop, I would guess, between highlight and shadows. So I don't really see how you can compare camera sensor DR without using exactly the same lens on both, which is usually impossible.
John
I don't doubt any of that. But I still think that to generalize and say that MF lenses have less flare and more DR than DSLR lenses is not true.
-
By the way, I do not think there are any secrets or conspiracies in the detector market. You could probably find the engineering specs for any extant chip package if you look in the right places.
I'd like to have a pointer to the official (not measured or reverse engineered) specs of the Canon sensors. Any links you are willing to share?
-
I would reckon that lens flare might have a much bigger impact than you might think. I did some tests (on film) a couple of years ago using two apparently identical Zeiss 80mm Planars (both silver C lenses ca 1971), off a tripod on a static subject outdoors under diffused overcast conditions (no deep shadows). I was using the 'Blad Pro Hood, so no direct light fell on the lens. One lens was T* coated, the other had the older single-coating. When printed, there was no difference in definition, but a noticeable difference in DR in favour of the T* - about half a stop, I would guess, between highlight and shadows. So I don't really see how you can compare camera sensor DR without using exactly the same lens on both, which is usually impossible.
John
John,
There's an interesting article on the imatest site that tests for veiling flare: - http://www.imatest.com/docs/veilingglare.html#intro (http://www.imatest.com/docs/veilingglare.html#intro) (thanks to Bill Janes for providing that link on the other thread - Dynamic Range and DXO).
The impression I get is that prime lenses will tend to have less glare than zooms, and that the removal of any filter, such as the usual protective UV filter, may help reduce glare. I always used to buy a UV filter with a new lens and have it permanently attached to protect the lens from scratches, something which amateurs tend to do because their lenses are so precious, like jewels.
However, for the past few years I've adopted the practice of never having a filter attached unless I need one for a particular effect, such as an ND filter or polarizer when photographing rivers and waterfalls. I believe that modern lens coatings are so tough and hard there's really no need at all for a protective filter.
It's possible that DXO results with regard to DR might be influenced slightly by the 'veiling glare' characteristics of the particular lens attached to the camera under test. However, it is very unlikely that DXO would test any camera using a 1971 lens with a single coating. As I understand, DXO either buy or rent the cameras they test, and I imagine they would use the best quality, modern, standard prime with each camera they test.
Towards the end of the imatest article there's a short list of some test results for a handful of lenses which are all zooms, with the exception of one prime, the Canon TS-E 90/2.8 which gets the best score.
It so happens I own a copy of the TS-E 90. I'm tempted to do a DR comparison with my 5D, comparing the TS-E 90 with the Canon 24-105 zoom at 90mm, to see just how significant any DR differences might be. I'd like to do it right now, today, but I can't justify spending the rest of the day on such a project when I have so many other urgent tasks to do, such as assembling dining room chairs which were delivered in cardboard cartons with an allen key and a set of 10 bolts for each chair. It takes a fair amount of time manually screwing 60 bolts with washers, using an allen key. There are 6 chairs and the holes don't seem perfectly aligned.
The reason I'm taking the time to write this is for a bit of mental stimulation.
-
I'd like to have a pointer to the official (not measured or reverse engineered) specs of the Canon sensors. Any links you are willing to share?
My apologies for being unclear. I was using "chip package" to specify a detector and support electronics marketed to OEMs. I had no trouble finding Dalsa's specs for basic things like well capacity, QE, etc. God knows what happens once raw processing is going on in the actual camera/computer, but it is nice to know we are going in there with reasonable specs.
Do we really need chip level specs? Photographers seem to want things that are at a higher level. Smooth skies and shadows as opposed to SNR, exposure latitude as opposed to DR, that sort of thing. The cameras are proprietary systems that deliver data to sastify what the manufacturers think we want. If you like the pictures, I suppose whatever is going on to massge the pixel data is good.
Hey, film is grossly nonlinear, has very limited DR, and is subject to variable treatment at every step of processing - but we tend to love what it shows. Eye of the beholder and all that.
-
I believe that lens flare will actually contribute to an effective INCREASE in usable dynamic range of the scene.
Veiling glare will lighten the shadows of a high DR image and make the image easier to print, but it would be better to use tone mapping to achieve this effect. With truly HDR images veiling glare limits the DR as explained in this PDF (http://graphics.stanford.edu/papers/glare_removal/glare_removal.pdf)from Stanford University. Deconvolution can remove some of the glare at the expense of increasing noise. An occlusion mask is another way to address the problem. The article gives some useful references. John McCann has an excellent 1 hour lecture on the topic on Google Lectures (http://www.youtube.com/watch?v=ALfiTDYLtAQ).
Bill
-
Hi,
No there is no really good reason to assume that, except possibly that MF lenses used to be more conservative designs. Not very fast, the zooms are few and have a narrow range. Premium DSLR lenses used to be quite fast.
Best regards
Erik
I don't doubt any of that. But I still think that to generalize and say that MF lenses have less flare and more DR than DSLR lenses is not true.
-
I feel that one key aspect in the perception of DR is the way the raw converters handles local contrast.
We know for a fact that both MFDB and high end DSLR have too much DR for a pleasing rendering without some tweaking. Images look flat when you attempts to map say 11 stops of DR on an 8 bit display.
So you need to work with local contract to generate pleasing results.
We know that the highlight/shadow sliders ot leading raw converters do play with local contrast, to the extend that they sometime generate halos similar to what one gets when using Photomatix.
How much of this is automated when you open a P65+ file in C1 is a topic on which I would be interested in getting more information.
Cheers,
Bernard
Bernard,
You can work with local contrast in various parts of the image to get "pleasing contrast" while still portraying the full DR the sensor and LR/Capture/Photoshop can deliver. The technique of developing the local contrast is important. I'm not too worried about what maps on the display, because even displays in the USD 1500 range don't show the full gamut or resolution of my printer (Epson 3800). The display is well colour-managed so it's reliable for overall colour balance, colour rendition within its gamut and luminosity, but that's about it. The bottom line for me is what comes out of the printer.
If you are using Lightroom, the local contrast enhancement of the dark tones is very well managed between the Fill and Blacks sliders. It's one of the most effective uses of that program. And Recovery is also very good for recovering highlights provided at least one channel has data. So this helps a lot with the high-end DSLRs. For raw files using the IIQ format of a Phase P40+ back, in Capture-1 version 5.1 there are shadow and highlight sliders. They do open-up shadows and tame highlights. I'm not nearly as experienced using this program as I am in using LR, but so far, this is one toolset of Capture-1 which doesn't turn me on very much. I find the effects are not "local" enough - too broadcast, unlike what I can get from LR. Nothing is really "automated" in Capture-1 if you don't want it to be. You can load the files into the program with everything set to zero or neutral and create your own "recipes" for various combinations of presets you want to apply to an image, or batch process.
-
I see your point Ray. I presume the interpolation of the larger files reduce noise and therefore increase DR.
And in the DxO "Print" condition (8 MP, 300 dpi, 8x12") is way beyond a healthy noise reduction on the P65 60 MP while it does wonders on an FF 12-24MP file. Nevertheless if the P65, D3x, D3s are compared in "Print"condition and at tonal range (so more linear) the P65 shows something of the difference in the graphs that it will give in print. Same for color depth. At lower ISO values of course.
If DxO added an extra "Large Print" condition to the existing one, say 15-20MP, 600 dpi, A2 size, the difference could be more significant.
The "Print" reference is a good idea in itself, the noise reduction on downsampling correct, but the limited print size/quality in favor or FF, APS and 4/3 and too low for MF.
met vriendelijke groeten, Ernst Dinkla
Try: http://groups.yahoo.com/group/Wide_Inkjet_Printers/ (http://groups.yahoo.com/group/Wide_Inkjet_Printers/)