Grant808 wrote:
Good points from everyone so far! One factor that may increase the effect of diffraction is the gapless microlenses. The more discrete data gathered from older sensors with larger gaps are just less likely to show the diffraction, which is always there for a given lens and settings.
I'm afraid that this is simply not going to be the case. Images don't line up with only vertical and horizontal components that neatly agree to not spill across photosite boundaries.
pahrens wrote:
So really it doesn't matter a whole lot that this sensor is so much denser than older models?
It might matter, but not in terms of diffraction blur.
For example, if you have a very high quality lens and shoot with extreme care you might get some tiny increment of increased system resolution if you select the optimum aperture. And you might get slightly better gradients with more, smaller photosites.
But the difference is really, really minor when it is visible at all. Unless you have very high quality lenses, shoot with extreme care from a tripod, and do very precise and careful post-processing, make very large prints, and then inspect them very closely...
... it is almost certain that you'll never notice any difference at all.
If you never print larger than letter size and/or you only post jpgs online or send them via email, this whole discussion is entirely pointless in terms of the quality of your photographs.
gdanmitchell wrote:
I'm afraid that this is simply not going to be the case. Images don't like up with only vertical and horizontal components that neatly agree to not spill across photosite boundaries.
Dan
Yeah, they don't 'like up' as you say. Lining up has nothing to do with what I am saying. The lack of sampling via the gaps makes diffraction effects less likely to show.
Apologies for the typo. I generally write my responses here very quickly and rarely proofread them with the same care that I would use when doing other types of writing.
I have corrected "like" to read "line."
But more to the point, this has everything to do with the point I was addressing. There is a mistaken idea that there is a relationship between the size of the photosites (or the gap between them) and the "amount" of diffraction in the image as the lens is stopped down - e.g. "If I have a camera with more photosites I'll have to open up more than I would on a camera with lower photosite density to as to avoid diffraction problems."
This is simply dead wrong. A more dense arrangement of photosites or reducing the gaps between them has essentially no effect on the amount of diffraction in the image. The photosite density increase can result in a more accurate rendition of whatever hits the sensor. (Yes, the 18MP sensor will produce a more accurate image of the diffraction blur than the 12MP sensor, though this is not a significant issue.) Decreasing the gaps between photosites without changing the density of photosites can increase the ability of the sensor to capture light and improve the signal to noise ratio, but it also has not sensible effect on the capture of diffraction blur. (Keep in mind that the image of every pixel in your capture is the result of interpolating color data from multiple photosites anyway.)
Dan
Grant808 wrote:
Yeah, they don't 'like up' as you say. Lining up has nothing to do with what I am saying. The lack of sampling via the gaps makes diffraction effects less likely to show.
gdanmitchell wrote:
Apologies for the typo. I generally write my responses here very quickly and rarely proofread them with the same care that I would use when doing other types of writing.
I have corrected "like" to read "line."
But more to the point, this has everything to do with the point I was addressing. There is a mistaken idea that there is a relationship between the size of the photosites (or the gap between them) and the "amount" of diffraction in the image as the lens is stopped down - e.g. "If I have a camera with more photosites I'll have to open up more than I would on a camera with lower photosite density to as to avoid diffraction problems."
This is simply dead wrong. A more dense arrangement of photosites or reducing the gaps between them has essentially no effect on the amount of diffraction in the image. The photosite density increase can result in a more accurate rendition of whatever hits the sensor. (Yes, the 18MP sensor will produce a more accurate image of the diffraction blur than the 12MP sensor, though this is not a significant issue.) Decreasing the gaps between photosites without changing the density of photosites can increase the ability of the sensor to capture light and improve the signal to noise ratio, but it also has not sensible effect on the capture of diffraction blur. (Keep in mind that the image of every pixel in your capture is the result of interpolating color data from multiple photosites anyway.)
No apologies needed. Though I'm sure we're not going to agree about this... As a Sigma/Foveon shooter, I am all too aware what Bayer interpolation is doing and how the colors are not even sampled in the exact spaces...and that might really make this whole sub-discussion moot.
Yet in theory you will lose more of the transitional data in the gapped condition than non-gapped data. The transitional data is where the diffraction will show. Lets use a straight diagonal line transition at say 31 degrees as an example. The gapped sensor will occasionally lose the transition data as you sample data on either side of the line. Sometimes it will land smack dab in the middle of the pixel, and sometimes run between...and there will be the diffraction effect missed. That's the data that will be more discrete. The gapless sensor will not miss any of that data along the line and all the diffraction info will be incorporated into the Bayer shake-n-bake.
I don't believe I can explain it any better...well, without a picture.
Grant808 wrote:
Yet in theory you will lose more of the transitional data in the gapped condition than non-gapped data. The transitional data is where the diffraction will show. Lets use a straight diagonal line transition at say 31 degrees as an example. The gapped sensor will occasionally lose the transition data as you sample data on either side of the line. Sometimes it will land smack dab in the middle of the pixel, and sometimes run between...and there will be the diffraction effect missed. That's the data that will be more discrete. The gapless sensor will not miss any of that data along the line and all the diffraction info will be incorporated into the Bayer shake-n-bake.
I don't believe I can explain it any better...well, without a picture. ...Show more →
I should probably just butt out here, but I can't resist a good diffraction argument...
Are you saying that diffraction creates features varying over a distance of less than a pixel width (the transitional areas of which you speak)? My impression is that diffractive blur will *increase* the spatial extent of an image feature - that's kind of what blurring means - and render it more likely to be sampled by multiple pixels rather than less.
I agree that gaps in the photosite array do allow for details to be missed that would be captured by a gapless array. But such details are almost by definition less than a pixel width in spatial extent. This means that as diffraction effects become more pronounced and their spatial extent increases they would be ever less likely to be unsampled in the region of the photosite gaps.
vachss wrote:
I should probably just butt out here, but I can't resist a good diffraction argument...
Are you saying that diffraction creates features varying over a distance of less than a pixel width (the transitional areas of which you speak)? My impression is that diffractive blur will *increase* the spatial extent of an image feature - that's kind of what blurring means - and render it more likely to be sampled by multiple pixels rather than less.
I agree that gaps in the photosite array do allow for details to be missed that would be captured by a gapless array. But such details are almost by definition less than a pixel width in spatial extent. This means that as diffraction effects become more pronounced and their spatial extent increases they would be ever less likely to be unsampled in the region of the photosite gaps.
+1 Not to mention the effect of the AA filter that takes the light that would have fallen on one sensel and spreads it out to (at least, according to Canon's description) 4 sensels each getting ~25% of the original light. The light that might fall in the cracks will also be spread out because the AA filter is before the filter/microlens array.
So the diffraction blur has been created before any light hits the light sensitive surface (so the nature of that surface - pixel size/grain size etc - does not affect it, only how it's recorded) and then it's spread out some more by the AA filter and THEN it gets to the filter/lens array. No wonder our images need PP
Of course it's going to become "visible" sooner on the higher resolving sensor. The higher resolving sensor will resolve more delicate detail that will be visibly affected by diffraction. There will be the same level of diffraction on the lower resolving sensor, but with no detail to be marred...nobody knows. Kind of like the old question about a tree falling in the woods:
"Is there still diffraction if there is no image detail to be lost to it?"
RDKirk wrote:
Of course it's going to become "visible" sooner on the higher resolving sensor. The higher resolving sensor will resolve more delicate detail that will be visibly affected by diffraction. There will be the same level of diffraction on the lower resolving sensor, but with no detail to be marred...nobody knows.
That's why I believe that in some cases many people are reporting that the 7D "is soft."
Depending on the aperture used in a given image, they are possibly starting to "see" diffraction far sooner than with other bodies when pixel-peeping at 100%.
I'm absolutely convinced that many, many of the famiiar reports that "the new higher photosite density camera doesn't perform as well as the previous model" posts can be traced to precisely this issue, namely that folks simply forget that when they look at 100% crops from cameras with different numbers of photosites it is equivalent to inspecting negatives under different levels of magnification.
A 100 x 100 section of a 18MP image comprises a much smaller section of the overall image than a 100 x 100 section of a 8MP image. It is entirely normal and expected that if you make exposures using the same lens and aperture on the two systems and then compare at 100% that you will see a "softer" image from the "sharper" 18MP sample.
Dan
cameron12x wrote:
That's why I believe that in some cases many people are reporting that the 7D "is soft."
Depending on the aperture used in a given image, they are possibly starting to "see" diffraction far sooner than with other bodies when pixel-peeping at 100%.
vachss wrote:
I should probably just butt out here, but I can't resist a good diffraction argument...
Are you saying that diffraction creates features varying over a distance of less than a pixel width (the transitional areas of which you speak)? My impression is that diffractive blur will *increase* the spatial extent of an image feature - that's kind of what blurring means - and render it more likely to be sampled by multiple pixels rather than less.
I agree that gaps in the photosite array do allow for details to be missed that would be captured by a gapless array. But such details are almost by definition less than a pixel width in spatial extent. This means that as diffraction effects become more pronounced and their spatial extent increases they would be ever less likely to be unsampled in the region of the photosite gaps.
I think you're misunderstanding me. I'm calling the transition line a change in the subject matter in color or tone. Like a line between red and green or light green and dark green...etc. Edge conditions that will show diffraction regardless of wether it is sub pixel or multi pixel width. The higher pixel density showing diffraction earlier was not in contention IMO...just the gap vs. gapless microlenses.
"Sub-pixel width diffraction" (an astonishingly unimportant and rare subset of the topic we are discussing) would typically spill to adjacent pixels anyway since it is extremely unlikely that such sources would be a) a single pixel tall/wide, b) perfectly centered over the middle of a photosite, c) likely part of larger structures that are not perfectly linear and perfectly vertically/horizontally aligned. Such features would also be "blurred" by the AA filter.
Dan
Grant808 wrote:
I think you're misunderstanding me. I'm calling the transition line a change in the subject matter in color or tone. Like a line between red and green or light green and dark green...etc. Edge conditions that will show diffraction regardless of wether it is sub pixel or multi pixel width. The higher pixel density showing diffraction earlier was not in contention IMO...just the gap vs. gapless microlenses.
AJSJones wrote:
+1 Not to mention the effect of the AA filter that takes the light that would have fallen on one sensel and spreads it out to (at least, according to Canon's description) 4 sensels each getting ~25% of the original light. The light that might fall in the cracks will also be spread out because the AA filter is before the filter/microlens array.
So the diffraction blur has been created before any light hits the light sensitive surface (so the nature of that surface - pixel size/grain size etc - does not affect it, only how it's recorded) and then it's spread out some more by the AA filter and THEN it gets to the filter/lens array. No wonder our images need PP ...Show more →
Well, that does point out that diffraction my only really be noticable when its effect is stronger or larger than the AA blurring.
gdanmitchell wrote:
"Sub-pixel width diffraction" (an astonishingly unimportant and rare subset of the topic we are discussing) would typically spill to adjacent pixels anyway since it is extremely unlikely that such sources would be a) a single pixel tall/wide, b) perfectly centered over the middle of a photosite, c) likely part of larger structures that are not perfectly linear and perfectly vertically/horizontally aligned. Such features would also be "blurred" by the AA filter.
So basically having a higher pixel density is not going to hinder the image quality any way over a lower one? I'm imagining if you resample the image down and compared with the less dense sensor they will look basically the same?
So if 18mp was sampled down to 10mp and compared with a 40D you shouldn't be able to tell the difference?
Or have I completely lost the point here?
pahrens wrote:
So basically having a higher pixel density is not going to hinder the image quality any way over a lower one? I'm imagining if you resample the image down and compared with the less dense sensor they will look basically the same?
So if 18mp was sampled down to 10mp and compared with a 40D you shouldn't be able to tell the difference?
Or have I completely lost the point here?
You got the point
Actuall, downsampling from 18 MP to 10 MP will give better results than the 40D. Diffraction will remain the same, but the anti-aliasing filter and de-bayer algorithms will work on a smaller scale in the image taken with a 18 MP sensor.
pahrens wrote:
So basically having a higher pixel density is not going to hinder the image quality any way over a lower one? I'm imagining if you resample the image down and compared with the less dense sensor they will look basically the same?
So if 18mp was sampled down to 10mp and compared with a 40D you shouldn't be able to tell the difference?
Or have I completely lost the point here?
That is not necessary a valid conclusion.
It is true that having higher photosite density will not create a "worse" image due to diffraction blur. Depending upon the aperture, the image will be at least as good and potentially a bit better in this regard on the higher MP sensor.
But diffraction blur is by no means the only issue related to image quality. If the camera/sensor manufacturer doesn't find ways to deal with the issue it is theoretically possible that dynamic range might decrease or that noise levels might increase. Despite this theoretical consideration, there is no evidence that continuous increases in DSLR photosite density have increased image noise or decreased dynamic range. In fact, if anything, the opposite is true.
Your "downsizing" notion would introduce some additional variables that do decrease image quality when you downsample. However, an interesting question would be whether a print of a given size from a 10MP sensor would look the same as a print from a 18MP sensor. Of course, if you mostly take your photos to online jpg output or only to small prints it doesn't matter one bit.
Lining up has nothing to do with what I am saying. The lack of sampling via the gaps makes diffraction effects less likely to show. 
Though I'm sure we're not going to agree about this... As a Sigma/Foveon shooter, I am all too aware what Bayer interpolation is doing and how the colors are not even sampled in the exact spaces...and that might really make this whole sub-discussion moot.