p.2 #1 · Ming Thein on pixel density and sharpness
Diffraction is also something to consider, but not mentioned in the article. The NEX-7 with 24 mpixels APS-C is already diffraction limited at f8. Anything with higher pixel density than the NEX-7 is going to be mostly useless.
p.2 #2 · Ming Thein on pixel density and sharpness
briantho wrote:
Diffraction is also something to consider, but not mentioned in the article. The NEX-7 with 24 mpixels APS-C is already diffraction limited at f8. Anything with higher pixel density than the NEX-7 is going to be mostly useless.
I agree with your first sentence and second sentence, but if you believe the third sentence please read this.
p.2 #3 · Ming Thein on pixel density and sharpness
I think I know what he means. Downrezz 36 mp to 24 mp, and you get more detail than a 24 mp native size. But the 24 mp will look sharper/crisper, perceptively.
But of course when you downrezz, you need to sharpen too
Btw, regarding diffraction, it depends on the sensor size, and output. A 36 mp FF sensor is not more diffraction limited than a 24 mp sensor.
Well, I'm no expert, but the argument that motion blur and focus error is the most common reason for unsharp photos, and therefore you shouldn't care about diffraction, is just moronic imho. I take care to focus correctly (manually) and to keep the camera steady, I wouldn't knowingly want to introduce another image degrading factor like diffraction.
p.2 #5 · Ming Thein on pixel density and sharpness
briantho wrote:
Well, I'm no expert, but the argument that motion blur and focus error is the most common reason for unsharp photos, and therefore you shouldn't care about diffraction, is just moronic imho. I take care to focus correctly (manually) and to keep the camera steady, I wouldn't knowingly want to introduce another image degrading factor like diffraction.
I think you're confused. Did you read the posts I linked?
p.2 #6 · Ming Thein on pixel density and sharpness
Jeff Kott wrote:
I think you're confused. Did you read the posts I linked?
Well not thoroughly
They talk about different other sources of image blur, and "how digital sensor resolution is almost impossible to characterize", resulting in worrying about diffraction is a waste of your time...
p.2 #7 · Ming Thein on pixel density and sharpness
briantho wrote:
Well, I'm no expert, but the argument that motion blur and focus error is the most common reason for unsharp photos, and therefore you shouldn't care about diffraction, is just moronic imho. I take care to focus correctly (manually) and to keep the camera steady, I wouldn't knowingly want to introduce another image degrading factor like diffraction.
Diffraction issues are a red herring in terms of increasing pixel count, because the higher resolution sensor is simply resolving the blur more. In other words, a higher megapixel sensor (assuming same sensor size) may not have a resolution advantage up to a certain f stop, because of diffraction, but it will never have a resolution disadvantage compared to a lower megapixel sensor at any aperture.
For example (and I'm making these numbers up,) a 100mp FF sensor may be diffraction limited at f2.8 and up, but a 12mp sensor won't have higher resolution from f2.8 and up, so there's no disadvantage to the 100mp sensor. Of course, buying a 100mp camera that only has a resolution advantage from f2.8 and wider may not be worth it to some, so it may start to be an issue of diminishing returns.
p.2 #8 · Ming Thein on pixel density and sharpness
I am on the high Mp side, along with Zeiss, cameras companies, the sensor designers and the physicists. I also harbour a longtime suspicion of the alarmism spread about diffraction, which is a gradual onset effect on final image quality, and I suspect there are a huge number of images that would be much improved had the user simply added a stop, if depth was the desired outcome.
Honestly, some people seem to think makers are hoodwinking people each time a higher density sensor is released. Note also that while technique is always important, no hue and cry has erupted for better IS/VR/IBIS now that we have 3-4 micron sensors? If anything fewer use tripods or support.
I agree Ed, and it really amounts to something quite separate - the visual perception (esp. on screen) which is a human visual system function emanating from different densities, hence all the voodoo downsizing recipes on the still current thread. People need it to 'look sharper', not be sharper, or carry more resolution.
p.2 #9 · Ming Thein on pixel density and sharpness
Thanks Douglas, that explains why my statement "Anything with higher pixel density than the NEX-7 is going to be mostly useless" was wrong. I'm changing it to:
Anything with higher pixel density than the NEX-7 is going to be of questionable added benefit.
p.2 #10 · Ming Thein on pixel density and sharpness
I think that there are uninformed consumers out there who think that the ONLY thing that matters about a camera is how many megapixels it has. (Not the lens quality, not the high ISO ability, not usability things like fast focusing, just the megapixels.)
Then there are these people online, who want to sound sophisticated, so they spin the opposite story that more pixels make the image quality WORSE.
Regarding resolution, the more pixels the camera has, the more real resolution should be recorded in the final image. This may be subject to diminishing returns. Doubling the number of pixels may not double the resolution. It may only increase by 90%. It might only increase by 20% if the lens is really bad or is already stopped down way beyond the diffraction limit. But it will always record more data and more resolution.
p.2 #11 · Ming Thein on pixel density and sharpness
briantho wrote:
Thanks Douglas, that explains why my statement "Anything with higher pixel density than the NEX-7 is going to be mostly useless" was wrong. I'm changing it to:
Anything with higher pixel density than the NEX-7 is going to be of questionable added benefit.
IF you're using a lens not designed to resolve that much sampling.
p.2 #12 · Ming Thein on pixel density and sharpness
I tend to think of it like this ...
It will capture & distinguish the smallest amount of detail the lens can project, then beyond that it will simply subdivide that smallest projected detail, if it is outresolving the projection. If the downrezzing is compiling subdivided detail, then it is essentially re-assembling it. If the downrezzing is compiling the smallest projected detail, then it is squashing it ... per the algorithm being used to contend with perceived contrast/sharpness in the process.
Absolutely no scientific basis for this ... just seems to stand to reason.
And, diffraction is an optical property of the projection, just like detail is an optical property of the projection. So whether you are sub-dividing the diffraction's loss of contrast or the detail ... I think that the pixel size, re: diffraction is more of an issue of the fact that such smaller format sensors are using smaller format lenses, with corresponding smaller apertures (which are a relative, not absolute size). The diffraction being attributed to small pixels is more of a corollary symptom, where the actual cause of the diffraction is the small absolute aperture size, not the small pixel.
p.2 #13 · Ming Thein on pixel density and sharpness
briantho wrote:
Thanks Douglas, that explains why my statement "Anything with higher pixel density than the NEX-7 is going to be mostly useless" was wrong. I'm changing it to:
Anything with higher pixel density than the NEX-7 is going to be of questionable added benefit.
That may be true for ya, although it depends on the lens. Also, the higher the pixel density, the less the need for AA filters. For aps-c, I think 36mp would still be fine, as it will cut down the need for an AA filter, and you'll still get resolution benefits from wide open to mid aperture.
p.2 #14 · Ming Thein on pixel density and sharpness
Toothwalker wrote:
If we assume a superior lens and technique, the original with the lower pixel count will be soft when there is an AA filter, or messed up when there is no AA filter. I would always pick the original with the largest pixel count as a starting point, provided the camera had a decent AA filter in place.
So let's take the D800, which has an AA filter. Are you saying that if you downsize a 36MP photo to a 35MP image, it will be sharper than a native 35MP camera would be? Sharper in terms of resolution or perception? I think Ming Thein is only talking about perception.
p.2 #15 · Ming Thein on pixel density and sharpness
Ming Thein is sometimes alarmingly cavalier about the details of how optics and sensors work. In this case he seems unconcerned that his proclamations fly in the face of basic information theory.
Looking at his article, I think his misconception starts here:
“Let’s take an example: if you have a scene shot with 12 (8.45um pixel pitch) and 36 MP (4.88um pixel pitch) cameras, and both are equally motion-blurred. The degree of camera shake may be less than a pixel on the 12MP camera – let’s say 5um for argument’s sake, which will still produce a critically sharp image.”
It’s not remotely true that an optimum sharpness is reached as long as the image movement on the sensor during exposure is kept below the pixel pitch. Any movement immediately degrades sharpness, and it becomes practically relevant well below the pixel pitch.
(I know he says ‘critically sharp’, not optimally sharp, but the rest of the article suggests that by ‘critically sharp’ he means ‘as good as it gets’.)
I have a rule of thumb, somewhat unfair, but useful: if someone advocates cameras without antialiasing filters, but can’t sensibly explain why they would want that, they probably don’t know what they’re talking about. Can anyone point me to where Ming Thein explains why he shoots with a D800E?
philip_pj wrote:
I also harbour a longtime suspicion of the alarmism spread about diffraction, which is a gradual onset effect on final image quality, and I suspect there are a huge number of images that would be much improved had the user simply added a stop, if depth was the desired outcome.
Agreed. Walter E Schön had a nice demonstration of this at the Rodenstock booth at this year’s Photokina. He said the demo would eventually be published on the Rodenstock website, but I haven’t seen it there yet.
p.2 #16 · Ming Thein on pixel density and sharpness
It's always the combination of original image resolution and magnification to the final viewing size that counts the most. You can look at a 16 x 20 print of a 4x5 LF image taken at f/64, and be absolutely blown away by the apparent sharpness - despite the "terrible" diffraction that "must be" associated with an image taken at f/64. It's all ratios...
p.2 #17 · Ming Thein on pixel density and sharpness
jcolwell wrote:
It's always the combination of original image resolution and magnification to the final viewing size that counts the most. You can look at a 16 x 20 print of a 4x5 LF image taken at f/64, and be absolutely blown away by the apparent sharpness - despite the "terrible" diffraction that "must be" associated with an image taken at f/64. It's all ratios...
+1
f64 on a 4x5 is probably a wee bit different (absolute aperture size) than f8 on a 4/3.
p.2 #18 · Ming Thein on pixel density and sharpness
jcolwell wrote:
It's always the combination of original image resolution and magnification to the final viewing size that counts the most. You can look at a 16 x 20 print of a 4x5 LF image taken at f/64, and be absolutely blown away by the apparent sharpness - despite the "terrible" diffraction that "must be" associated with an image taken at f/64. It's all ratios...
RustyBug wrote:
+1
f64 on a 4x5 is probably a wee bit different (absolute aperture size) than f8 on a 4/3.
Could be a wee bit, but I don't think the difference is large. After all, f-stop = (focal length) / (apparent aperture diameter as viewed through front element). Many f/32, f/45, and f/64 LF lenses had smaller front elements than many current M4/3 lenses...
p.2 #19 · Ming Thein on pixel density and sharpness
Actually, Ming is spot on - if you make leeway for the: -"as normal photographers shoot and develop their images" statement, also spiced up with the words "perceptual sharpness".
It is not universally true, and it's ABSOLUTELY untrue from both a technical and a theoretical PoV.
Modern raw conversion softs are geared towards maximizing "detail per pixel position on the sensor". As owners of AA-less cameras know, this produces false detail and edge dither. This improves the perceived sharpness and "detail" in the image, and most of the time (until you hit a moire pattern or other pattern-shaped errors) the viewer will be none the wiser to the fact that all the small grit that he/she perceives as "detail" wasn't actually there in reality in front of the camera.
This improves "detail per pixel" in a highly synthetic way, but in a way that we've spent large sums of soon twenty years of development budgets on to make "visually attractive".
.................
The other part is the sad state of almost all resampling engines out there. I'd like to think I know most available software out there, and not ONE of them does resampling (scaling) in any way close to ideal or correct. There's two parts to this, the sampling algorithm and the concept of gamma.
Sampling algorithm:
There's a good reason some people say that the factor 1.50-1.66 is ideal in "steps" for scaling. In most normal softs, the standard resampling algorithms are to smooth when using a lower factor, and add to much sharpening (actually aliasing!) when using larger factors.
Downsampling from 36>24 MP is a factor of 1.22. That will make the image lose contrast in pixel>pixel boundaries. 36>16 is a factor of 1.5, and here the standard BiCubic starts to work better. 36>1 (web size) is a factor of 6 and will cause serious aliasing of sampling points, moire and strange patterns will be visible.
And that's only to save processing time. Even though a well written better algorithm will scale a 36MP image to 24MP in less than 1/10s on a good computer, the writers of software almost always targets processing time on the average 2-year old computer handled by a dork that wouldn't see the IQ difference anyway.
And then Gamma:
When resampling in a gamma-corrected color space, all the values are skewed. Blend two values together by averaging, and you get the wrong result.
The only correct way to apply the resampling algorithm is on linear data (gamma = 1.0). Then the values that the algorithm works on will provide the correct "averages" numerically, and internal contrast and color will scale correctly.
The gamma also has a very profound effect on sharpening - what we call "capture sharpening" is often made in linear space, while "output sharpening" is made in gamma-corrected space. And the output sharpening is tuned to work well with the defects and faults already built in to the downsampled file, making the result slightly off even before it began.
...............
So - while he is basically correct in practice, that's just because the image data is handled incorrectly. Which of course isn't his fault, that's just the way things work right now.
When correctly handled, a downsampled image is always sharper than a 1:1 size raw file, as long as all the surrounding circumstances are equal. I even remember showing that with a D3x and a D3 many, many years back now it seems...
p.2 #20 · Ming Thein on pixel density and sharpness
Another great, informative post suede, thanks. You've mentioned these downsampling ratios before in the context of demosaicing, not just for resolution and color but also for noise reduction, which is important for getting a D800 image to match the theoretical per-area noise equivalence of a lower resolution sensor with a similar QE, like the D3s.
I think a hole in Ming's argument which he only partially addresses is how even the lower-native resolution image has to undergo downsampling as well on its way to its final rendering on the screen and print, which may nullify the pixel-level differences between sensors that he's observing on the screen. He mentions this his closing statements but I think most readers would probably gloss over the significance of it.
