Pixel size and image resolution
Moderators: rjlittlefield, ChrisR, Chris S., Pau
Pixel size and image resolution
There are a lot of threads in this forum regarding various equipment.Microscope objectives,tube lenses,macro lenses are thoroughly tested and this is good ,because knowledge is shared.
However there is a lack of testing regarding image sensors despite the fact it is the weakest chain link.(it is well known that for magnifications less than 10x, sensors of hundreds of million pixels are required to match the theoretical resolutions of microscope objectives and macro lens. So in general lenses out resolve sensors by far.)
This is a rough test I made recently comparing two cameras with the same lens. Sony A6000 (pixel size 3.9um) was tested against Canon 5D III (pixel size 6.25um).For comparison I used a canon 70-200L f4 at 200mm at F4 on both cameras.The cameras were on tripod focus was manual using camera's magnification aids and high speeds were used to avoid vibrations.
For my convenience I used highest resolution jpgs from both cameras.
1)Sony A6000 100%
2)Canon 5D III 100%
You can easily notice that the canon 70-200 has information that can't be recorded from 5Diii lower pixel density sensor.No matter I increase the magnification the sign is unreadable.
However there is a lack of testing regarding image sensors despite the fact it is the weakest chain link.(it is well known that for magnifications less than 10x, sensors of hundreds of million pixels are required to match the theoretical resolutions of microscope objectives and macro lens. So in general lenses out resolve sensors by far.)
This is a rough test I made recently comparing two cameras with the same lens. Sony A6000 (pixel size 3.9um) was tested against Canon 5D III (pixel size 6.25um).For comparison I used a canon 70-200L f4 at 200mm at F4 on both cameras.The cameras were on tripod focus was manual using camera's magnification aids and high speeds were used to avoid vibrations.
For my convenience I used highest resolution jpgs from both cameras.
1)Sony A6000 100%
2)Canon 5D III 100%
You can easily notice that the canon 70-200 has information that can't be recorded from 5Diii lower pixel density sensor.No matter I increase the magnification the sign is unreadable.
-
- Posts: 3416
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
In this case the feature size on subject is smaller than the sensor can resolve, given the fixed magnification (it was fixed, correct?). This can easily be calculated, assuming all sensors resolve in a theoretically predictable manner. Of course some sensors will resolve smaller details than other sensors, even if they have the same physical resolution, due to things like anti-aliasing, de-mosaicing algorithms, etc. Perhaps this is a useful test, ie how well does a sensor resolve relative to its theoretical limit, given that the optics feeding it are not limiting.
Edited to add:
Indeed this is of prime interest to me, since I have purchased 3 cameras (Pentax, Sony, and Canon) which offered pixel-offset resolution enhancement, yet did not live up to expectations.
Edited to add:
Indeed this is of prime interest to me, since I have purchased 3 cameras (Pentax, Sony, and Canon) which offered pixel-offset resolution enhancement, yet did not live up to expectations.
Ray, I've been meaning to ask you about those pixel-shifting results. On the Sony, you presented results which seemed to indicate that the pixel-shifting was generating false detail. I've been wondering about that, because I am considering buying a Sony A7riii and the pixel shift, if it worked, would eliminate my need for a monochrome camera (since the shift causes R, G, and B to be measured at every point in the image).
The Sony puts a mandatory 1 second delay between its four shifted shots. My Cognisys (but not my WeMacro) stacking rail often drifts slightly, and would might move by more than 3 microns over the course of a second. Might this have caused your problem? Do you remember your settling time setting? If it were long enough, the movement would eventually stop. So you could test this hypothesis by doing a stack with long settling times.
I've been very happy with the Olympus shifting algorithm, when there is no movement. The Sony and Pentax shifting algorithms don't actually increase resolution over the nominal resolution (rather, not using the shifting algorithms reduces the effective resolution to values significantly lower than the nominal one), but the Oly algorithm really does deliver a significant increase in resolution over the nominal sensor resolution value.
The Sony puts a mandatory 1 second delay between its four shifted shots. My Cognisys (but not my WeMacro) stacking rail often drifts slightly, and would might move by more than 3 microns over the course of a second. Might this have caused your problem? Do you remember your settling time setting? If it were long enough, the movement would eventually stop. So you could test this hypothesis by doing a stack with long settling times.
I've been very happy with the Olympus shifting algorithm, when there is no movement. The Sony and Pentax shifting algorithms don't actually increase resolution over the nominal resolution (rather, not using the shifting algorithms reduces the effective resolution to values significantly lower than the nominal one), but the Oly algorithm really does deliver a significant increase in resolution over the nominal sensor resolution value.
Yes it is the same magnification since the same lens is used in both cameras at the same distance from subject. To be honest I was expected a less dramatic difference since I used a zoom lens at maximum focal distance and wide open. Another consequence will be that using a prime lens for example the Canon 200mm F2 you will see no increase in resolution since the limiting factor is the sensor of 5DIII. Of course this is of minor importance in normal everyday photography , but of great importance in macro and micro photography.ray_parkhurst wrote:In this case the feature size on subject is smaller than the sensor can resolve, given the fixed magnification (it was fixed, correct?). This can easily be calculated, assuming all sensors resolve in a theoretically predictable manner. Of course some sensors will resolve smaller details than other sensors, even if they have the same physical resolution, due to things like anti-aliasing, de-mosaicing algorithms, etc. Perhaps this is a useful test, ie how well does a sensor resolve relative to its theoretical limit, given that the optics feeding it are not limiting.
Edited to add:
Indeed this is of prime interest to me, since I have purchased 3 cameras (Pentax, Sony, and Canon) which offered pixel-offset resolution enhancement, yet did not live up to expectations.
For example:How a printing nikkor 105mm will behave in a full frame sensor of 24mpx and in an APS sensor of same pixel count?I'm very curious to see such a comparison( I would expect to see striking differences even with a mitutoyo 5x)
-
- Posts: 1953
- Joined: Sat Oct 07, 2006 10:16 am
- Location: Bigfork, Montana
- Contact:
-
- Posts: 3416
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
Lou,Lou Jost wrote:Ray, I've been meaning to ask you about those pixel-shifting results. On the Sony, you presented results which seemed to indicate that the pixel-shifting was generating false detail. I've been wondering about that, because I am considering buying a Sony A7riii and the pixel shift, if it worked, would eliminate my need for a monochrome camera (since the shift causes R, G, and B to be measured at every point in the image).
The Sony puts a mandatory 1 second delay between its four shifted shots. My Cognisys (but not my WeMacro) stacking rail often drifts slightly, and would might move by more than 3 microns over the course of a second. Might this have caused your problem? Do you remember your settling time setting? If it were long enough, the movement would eventually stop. So you could test this hypothesis by doing a stack with long settling times.
I've been very happy with the Olympus shifting algorithm, when there is no movement. The Sony and Pentax shifting algorithms don't actually increase resolution over the nominal resolution (rather, not using the shifting algorithms reduces the effective resolution to values significantly lower than the nominal one), but the Oly algorithm really does deliver a significant increase in resolution over the nominal sensor resolution value.
I don't remember any control over the settling times, but it's been a little while now. I do know I took several shots of the same subject, and saw exactly the same thing each time, so I ruled out misalignment due to vibration as the cause.
I also used the Sony software to compose the images. Robert let me know that there are other software packages that might give better results, but unfortunately it was too late and I have had little time to go back and test the theory on my test shots. I can provide those to you if interested so you can try different software before you buy. My suspicion is the Sony algos were just not giving enough control over sharpening, which I believe is the problem. The images seemed over-sharpened at fine detail level. I saw the same issue with the 5DSR, but the RAW images were adjustable to, unlike the Sony ARQ.
Ray
Last edited by ray_parkhurst on Wed May 09, 2018 3:24 pm, edited 1 time in total.
I think the better color rendition due to pixel shift helps in resolutionLou Jost wrote:The Sony and Pentax shifting algorithms don't actually increase resolution over the nominal resolution (rather, not using the shifting algorithms reduces the effective resolution to values significantly lower than the nominal one), but the Oly algorithm really does deliver a significant increase in resolution over the nominal sensor resolution value.
Sony A7riii and Pentax K1ii on dpreview
-
- Posts: 2627
- Joined: Thu Jan 17, 2013 9:34 pm
- Location: United States
- Contact:
Re: Pixel size and image resolution
This is the best site by far for comparing all the different bodies with 3 or 4 different targets including resolution at all different ISO settings all side by side. You can even download the full size files to examine on your own.harisA wrote:.....
However there is a lack of testing regarding image sensors despite the fact it is the weakest chain link....
.
I believe they use a Sigma 70mm macro for all the different mounts.
https://www.imaging-resource.com/IMCOMP/COMPS01.HTM
Give it a try.
This is the Canon MKIII on the left and Sony A6000 at 100% pixel view, side by side.
-
- Posts: 3416
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
There is no change in "resolution" as measured in megapixels, but there is the potential for increased image resolution in terms of MTF. The Oly actually increases the MP resolution.lothman wrote:I think the better color rendition due to pixel shift helps in resolutionLou Jost wrote:The Sony and Pentax shifting algorithms don't actually increase resolution over the nominal resolution (rather, not using the shifting algorithms reduces the effective resolution to values significantly lower than the nominal one), but the Oly algorithm really does deliver a significant increase in resolution over the nominal sensor resolution value.
Edited to add:
Robert, that's a great site for comparisons! Thanks for posting it. Unfortunately I only see one mode for each camera, ie for the A7R3 there is no high res mode comparison, and for the 5DSR there is no fine detail mode. Or am I missing something?
Last edited by ray_parkhurst on Wed May 09, 2018 3:52 pm, edited 1 time in total.
but "resolution" is measured in pixel/distance not in number of pixels ;-)ray_parkhurst wrote: There is no change in "resolution" as measured in megapixels, but there is the potential for increased image resolution in terms of MTF. The Oly actually increases the MP resolution.
is this no increase of resolution?
-
- Posts: 3416
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
"resolution" has a lot of meanings.lothman wrote:but "resolution" is measured in pixel/distance not in number of pixels ;-)ray_parkhurst wrote: There is no change in "resolution" as measured in megapixels, but there is the potential for increased image resolution in terms of MTF. The Oly actually increases the MP resolution.
-
- Posts: 3416
- Joined: Sat Nov 20, 2010 10:40 am
- Location: Santa Clara, CA, USA
- Contact:
What I found with the A7R3 was that the pixel shift images "appeared" higher resolution, but most or all was due to a higher sharpening being done during the pixel shift compositing. Increasing the sharpening of the normal image gave almost the same result, but did not create the "donut" artifacts I saw with the pixel shifted image.
edited after Lou comment:
I think there is an over-sharpening happening in the compositing that is emphasizing the artifacts created during compositing process.
I saw a similar problem with the Pentax I bought (and returned) which has pixel shift. The process created a "pattern" that improved resolution but did not respond well to sharpening, and the Pentax process also gave little adjustability in sharpness adjustment. It may be that a A7R3 ARQ image, with reduced pixel-level (or perhaps "sub-pixel" level) sharpening may give a much better result.
edited after Lou comment:
I think there is an over-sharpening happening in the compositing that is emphasizing the artifacts created during compositing process.
I saw a similar problem with the Pentax I bought (and returned) which has pixel shift. The process created a "pattern" that improved resolution but did not respond well to sharpening, and the Pentax process also gave little adjustability in sharpness adjustment. It may be that a A7R3 ARQ image, with reduced pixel-level (or perhaps "sub-pixel" level) sharpening may give a much better result.
Isn't pixel shift more about improving colour than resolution? Each pixel in the final image gets a full set of (true) R, G and B values in the inputs - none interpolated from the Bayer matrix. This may have the effect of making the image look sharper (clearer), but resolution is not increased (AFAIK).