We got into a partial discussion surrounding the different "looks" generated by the different sensors we were comparing in the P65+ sneak peek thread (see here: http://forum.getdpi.com/forum/showthread.php?t=4333). A few folks commented on how the P25+ files with their "fatter" 9 micron pixels appeared to generate an image with more "pop." This is a fairly common reaction to 9 micron sensors from any of the manufacturers and most of us who look at a lot of digital files do see it, myself included. So I decided it might be worthwhile to start a thread about it where we can share some theories.
I say theories, because in trying to get my arms around the technical and mathematical aspects of cross-comparing many of the factors I believe contribute -- such as total DR, inter-pixel-contrast, Nyquist limits and lens performance -- I quickly realized I was going to have make too many assumptions to nail anything down in a concrete fashion. Moreover, I "see" what folks are talking about, so I know it is a "real" trait from that standpoint -- and can therefore look at it empirically. That conflicts with the mathematician-engineer in me, but is perfectly suited to my inner-artist, and since empirical is easier to compare visually, my inner-artist won out over my inner-mathematician . Here is an overview of some image comparisons I performed -- I used the series of "Across the Lake" test shots we already showed in full and crops from in the main thread.
Note these are my theories, observations and conclusions and presented here only to start the discussion -- I am *not* claiming them as facts.
For the first set of trials, I opened the three files that were as identical in framing as possible in C1 (4.5.1). My goals were to get the files as close to the same WB and overall color as possible, get them as close to the same exposure as possible, get as similar an inter-pixel appearance when viewed at 100% as possible and get the deepest blacks to the same point. This was attempted ALL in the raw converter, then the files were processed as 16-bit Profoto Tiffs for later working in Photoshop CS4 to finish the task. To accomplish the similar look, here is what was required in C1: I first matched WB as close as possible. The P25+ and P45+ were virtually identical when I was done, but the P65+ has a distinctly different color balance, so it only kind of close. Next I equalized local contrast (clarity) -- it is pretty obvious the P25+ has the most -- which required adjustments to the P45+ and P65+ files. In C1 I added 8 points clarity for the P45+ and 12 points for the P65+. Note that clarity also affects total contrast, and at this point most of the differences in perceived DR have evaporated from the files... Next I adjusted highlight exposures to the same level as the darkest image, in this case, the P25+. (It is important to note that none of the files had clipped highlights so I could have alternatively increased the two darker to match the lightest, P65+, but the slightly darker file showed better overall saturation, so I went that way.) I decreased the exposure on the P45+ file by 0.2 stop and on the P65+ by 0.4 stop. Finally I ratcheted the deepest black in the image to the same point in all three files. This required a minor levels adjustment to the each file -- I moved the black slider from zero to 6 on the P25+ file, zero to 4 on the P45+ file and zero to just 1 on the P65+ file. (Ironically, my gut tells me this indicates the P25+ may actually carry slightly MORE total DR than the other two backs!) At this point all of the images look very, very close save for a slight color difference in the P65+ and slight saturation differences (which I correct for in CS4 as explained below).
Once in CS4, I balanced saturation. I did it here because there was some differential saturation in the P65+ file I knew would get equalized by a vibrance adjustment, and C1 does not contain a vibrance slider as does LR/ACR. Here a 10 point bump in vibrance combined with a 10 point bump in saturation was given to the P65 file. The P45 got a 6 point bump in saturation only, and now all three files had pretty similar overall saturation. I then uprezzed ALL of the files to 24"x32" prints at 360 PPI. This required only a marginal 50% uprez on the P65+ file, but a significant 400% increase on the P25+ file. My goal here was to have ALL files uprezzed at least some amount, as we might do with each file in a real-world example for a larger print. The advantage of uprezzing to a print size of 24x32 at 360 PPI is it also serves double-duty as an identical print file for doing a 36"x48" print at 240 PPI.
The one thing that is very difficult to show online is how prints look comparatively. What I have found is that you can get a reasonable approximation of how final prints will compare by viewing the files side-by-side at 50% view in PS. This I did do and grabbed a screen-shot of all three "crops shown onscreen side-by-side. This is a full-screen cut off my 30" display, so is a large jpeg. I am linking to it instead of embedding it here, click on the thumbnail. You can easily tell which image is which by reading the titles. Note that these are *NOT* sharpened for output yet, but may show some jpeg compression artifacts:
~~~
A summary of my observations viewing all of these onscreen:
1) First, I feel I did get the 9 micron "look" in the other two files using my set of adjustments. IMO I would thus call the myth "busted," though admittedly I feel the 9 micron sensors deliver the better look without added work, so maybe there is something to the "magic" side.
2) You can definitely see the progressive additional detail as we move up in sensor resolution at 100%, but they are pretty slight and diminish to almost insignificant levels in the 50% view.
3) The P25+ has a propensity to generate visible moire in areas of repetitive fine detail like the mini-blinds in the windows seen in the screen capture. I conclude this tends to happen most often when the lens is sharper than the sensor, but also suspect the lens being sharper than the sensor is paramount to why the 9 micron sensors exhibit more inter-pixel contrast without added adjustments. Pick your poison.
4) The P45+ and P65+ 50% views are very, very close to my eyes, with an ever so slight advantage to the P65+ in the areas of really fine detail like the individual willow leaves, blades of tall grass and thinnest branches on the bushes and trees. Here the P25+ looks a bit softer for sure, but still resolves most of these same detail parts as well-defined individuals. I conclude here that the minor differences between the P45+ and P65+ may be because we are very near the limit of lens resolution already in the P45+.
Now for my summary after viewing actual prints.
1) In the 24x32 inch prints, the difference in discernible detail is IMO insignificant. In fact, I can only detect actual detail differences by viewing the print with a loupe. Using my reading cheaters at 12", I cannot see any significant differences -- really!
2) In the 36x48 inch prints... Here, with my nose in the prints, I can detect a slight loss of detail in the P25+ print -- it is very minor, but it is visible. However, it is not really noticeable unless you are comparing the P25+ directly to one of the others with your nose in the print. Between the P45+ and P65+, I see no difference. Again, if I use the loupe, I can see a slight edge to the P65+ print -- but even then it's a very tiny difference. From this observation I conclude the P25+ is starting to hit a limit on critical (very critical) detail at 36x48 inch print sizes. As photographers, I feel we are generally far more attuned to the minor nuances of detail than "normal" art viewing folk, so no doubt the P25+ could still go twice that large for most normal viewing situations.
3) Moire in the P25+ (not sure it carries through to other 9 micron pixel backs, but I assume it does) is a real issue and the user needs to learn/know how to deal with it. Note that the C1 software DOES contain a moire reducer (don't know about the others), but I did not run it for this test as it also kills very fine detail. My solution would be to run two conversions of the same file, one with and one without moire reduced, then layer them in CS and mask out the offending moire areas -- in this way the final image maintains the best from both conversions. Alternatively, if the uber-fine detail is not relevant to the image -- which in all honesty it often isn't -- then by all means save the masking hassle and use the moire tool (if available) during raw conversion.
A final thought: If you are regularly going to print larger than 24x36, then for sure the smaller pixel backs are probably worth looking at for their added ability to render very fine detail. However, understand you may become "lens limited" quickly and may need to consider an alternative camera system with digital-specific or high-resolution lenses to fully utilize the resolution the back is capable of.
Cheers,
I say theories, because in trying to get my arms around the technical and mathematical aspects of cross-comparing many of the factors I believe contribute -- such as total DR, inter-pixel-contrast, Nyquist limits and lens performance -- I quickly realized I was going to have make too many assumptions to nail anything down in a concrete fashion. Moreover, I "see" what folks are talking about, so I know it is a "real" trait from that standpoint -- and can therefore look at it empirically. That conflicts with the mathematician-engineer in me, but is perfectly suited to my inner-artist, and since empirical is easier to compare visually, my inner-artist won out over my inner-mathematician . Here is an overview of some image comparisons I performed -- I used the series of "Across the Lake" test shots we already showed in full and crops from in the main thread.
Note these are my theories, observations and conclusions and presented here only to start the discussion -- I am *not* claiming them as facts.
For the first set of trials, I opened the three files that were as identical in framing as possible in C1 (4.5.1). My goals were to get the files as close to the same WB and overall color as possible, get them as close to the same exposure as possible, get as similar an inter-pixel appearance when viewed at 100% as possible and get the deepest blacks to the same point. This was attempted ALL in the raw converter, then the files were processed as 16-bit Profoto Tiffs for later working in Photoshop CS4 to finish the task. To accomplish the similar look, here is what was required in C1: I first matched WB as close as possible. The P25+ and P45+ were virtually identical when I was done, but the P65+ has a distinctly different color balance, so it only kind of close. Next I equalized local contrast (clarity) -- it is pretty obvious the P25+ has the most -- which required adjustments to the P45+ and P65+ files. In C1 I added 8 points clarity for the P45+ and 12 points for the P65+. Note that clarity also affects total contrast, and at this point most of the differences in perceived DR have evaporated from the files... Next I adjusted highlight exposures to the same level as the darkest image, in this case, the P25+. (It is important to note that none of the files had clipped highlights so I could have alternatively increased the two darker to match the lightest, P65+, but the slightly darker file showed better overall saturation, so I went that way.) I decreased the exposure on the P45+ file by 0.2 stop and on the P65+ by 0.4 stop. Finally I ratcheted the deepest black in the image to the same point in all three files. This required a minor levels adjustment to the each file -- I moved the black slider from zero to 6 on the P25+ file, zero to 4 on the P45+ file and zero to just 1 on the P65+ file. (Ironically, my gut tells me this indicates the P25+ may actually carry slightly MORE total DR than the other two backs!) At this point all of the images look very, very close save for a slight color difference in the P65+ and slight saturation differences (which I correct for in CS4 as explained below).
Once in CS4, I balanced saturation. I did it here because there was some differential saturation in the P65+ file I knew would get equalized by a vibrance adjustment, and C1 does not contain a vibrance slider as does LR/ACR. Here a 10 point bump in vibrance combined with a 10 point bump in saturation was given to the P65 file. The P45 got a 6 point bump in saturation only, and now all three files had pretty similar overall saturation. I then uprezzed ALL of the files to 24"x32" prints at 360 PPI. This required only a marginal 50% uprez on the P65+ file, but a significant 400% increase on the P25+ file. My goal here was to have ALL files uprezzed at least some amount, as we might do with each file in a real-world example for a larger print. The advantage of uprezzing to a print size of 24x32 at 360 PPI is it also serves double-duty as an identical print file for doing a 36"x48" print at 240 PPI.
The one thing that is very difficult to show online is how prints look comparatively. What I have found is that you can get a reasonable approximation of how final prints will compare by viewing the files side-by-side at 50% view in PS. This I did do and grabbed a screen-shot of all three "crops shown onscreen side-by-side. This is a full-screen cut off my 30" display, so is a large jpeg. I am linking to it instead of embedding it here, click on the thumbnail. You can easily tell which image is which by reading the titles. Note that these are *NOT* sharpened for output yet, but may show some jpeg compression artifacts:
~~~
A summary of my observations viewing all of these onscreen:
1) First, I feel I did get the 9 micron "look" in the other two files using my set of adjustments. IMO I would thus call the myth "busted," though admittedly I feel the 9 micron sensors deliver the better look without added work, so maybe there is something to the "magic" side.
2) You can definitely see the progressive additional detail as we move up in sensor resolution at 100%, but they are pretty slight and diminish to almost insignificant levels in the 50% view.
3) The P25+ has a propensity to generate visible moire in areas of repetitive fine detail like the mini-blinds in the windows seen in the screen capture. I conclude this tends to happen most often when the lens is sharper than the sensor, but also suspect the lens being sharper than the sensor is paramount to why the 9 micron sensors exhibit more inter-pixel contrast without added adjustments. Pick your poison.
4) The P45+ and P65+ 50% views are very, very close to my eyes, with an ever so slight advantage to the P65+ in the areas of really fine detail like the individual willow leaves, blades of tall grass and thinnest branches on the bushes and trees. Here the P25+ looks a bit softer for sure, but still resolves most of these same detail parts as well-defined individuals. I conclude here that the minor differences between the P45+ and P65+ may be because we are very near the limit of lens resolution already in the P45+.
Now for my summary after viewing actual prints.
1) In the 24x32 inch prints, the difference in discernible detail is IMO insignificant. In fact, I can only detect actual detail differences by viewing the print with a loupe. Using my reading cheaters at 12", I cannot see any significant differences -- really!
2) In the 36x48 inch prints... Here, with my nose in the prints, I can detect a slight loss of detail in the P25+ print -- it is very minor, but it is visible. However, it is not really noticeable unless you are comparing the P25+ directly to one of the others with your nose in the print. Between the P45+ and P65+, I see no difference. Again, if I use the loupe, I can see a slight edge to the P65+ print -- but even then it's a very tiny difference. From this observation I conclude the P25+ is starting to hit a limit on critical (very critical) detail at 36x48 inch print sizes. As photographers, I feel we are generally far more attuned to the minor nuances of detail than "normal" art viewing folk, so no doubt the P25+ could still go twice that large for most normal viewing situations.
3) Moire in the P25+ (not sure it carries through to other 9 micron pixel backs, but I assume it does) is a real issue and the user needs to learn/know how to deal with it. Note that the C1 software DOES contain a moire reducer (don't know about the others), but I did not run it for this test as it also kills very fine detail. My solution would be to run two conversions of the same file, one with and one without moire reduced, then layer them in CS and mask out the offending moire areas -- in this way the final image maintains the best from both conversions. Alternatively, if the uber-fine detail is not relevant to the image -- which in all honesty it often isn't -- then by all means save the masking hassle and use the moire tool (if available) during raw conversion.
A final thought: If you are regularly going to print larger than 24x36, then for sure the smaller pixel backs are probably worth looking at for their added ability to render very fine detail. However, understand you may become "lens limited" quickly and may need to consider an alternative camera system with digital-specific or high-resolution lenses to fully utilize the resolution the back is capable of.
Cheers,