The lenses we use

[ray_parkhurst]

My point, Ray, is that this is possible because you put extra pixels under that feature. It is logically just the same as if you had a sensor with twice the pixel density, and then downsized it.

Sure, it is a similar result. But why does that result in more sharpness versus the non-downsized version? Rik's earlier assessment was that for monochrome subjects (like line targets...or silver coins) the bayer sensor should have essentially full resolution. My camera does not have an AA filter, so I can't blame it. It does still have its IR cut filter, so that may be causing some issues. Ultimately, fixing this degradation of 100% pixel detail is the "other end" of the resolution quest for me.

[ray_parkhurst]

I did not understand about same resolution with less pixels working for the same FoV. It must covers same FoV but with twice less pixels across side.
With 2X TC your projection size on sensor must be twice less and cover 4 time less pixels.

This is the correct analysis of the original images. But I cropped both images to the same FOV, then downsized the 2XTC image by 2x. The published images now both have same FOV and same number of pixels. Both are working at the same EA. Both were supplied similar information to the sensor, at least in the central part of the image where I did the crop.

Now, the process of 2x optical expansion and 2x digital downsizing doesn't add information to the image (at least not to the in-focus areas where I cropped), so any un-sharpness in one final image vs the other must be a result of incomplete information capture by the sensor.

edited to add: I sort of skirted the issue of increased DOF due to the smaller EA when the TC is added. This is an optical effect that is not undone when the image is downsized. You can see the TC images have better DOF than the non-TC, so look overall sharper, but the critical focus plane is not affected by the increased DOF.

[Justwalking]

Interesting, can be achieved exactly same result optically magnify by 10 and then dounsized digitally by 10 as at 1:1 on crop comparision?

[ray_parkhurst]

Interesting, can be achieved exactly same result optically magnify by 10 and then dounsized digitally by 10 as at 1:1 on crop comparision?

I'm not sure what happens to DOF if you magnify the image using a 10x objective with narrower NA than the primary objective. I'd guess the narrow DOF of the 10x objective will set the overall DOF.

[Lou Jost]

Ray, I think you are right, this is essentially doing what Pentax or Sony do in their pixel-shifting. By expanding the optical image, you are now taking full RGB data for each of the original pixels (whereas originally only R or G or B data were taken at any given pixel). So you are eliminating the resolution loss due to the Bayer filter. I think a higher magnification would be required to begin to show that there is more info in the aerial image.

But in any case, more pixels is better.

[ray_parkhurst]

I think a higher magnification would be required to begin to show that there is more info in the aerial image.

For sure, the 2xTC magnified image has more information than the non-TC image. But when you downsize by 2x, that information is lost.

Think of it in terms of your stereo. The signal coming from your CD player can convey lots of information, with wide dynamic range. The music contained therein has wide range of tonality and subtle gradients. Think of the music as the subject (coin, butterfly wing) and the, CD player as a good lens. The volume control is analogous to the magnification you're viewing at, and your ears are analogous to the camera sensor. If you turn up the volume control 6dB, you can hear quiet passages in the music that were below your threshold of hearing at lower volume. But then you add 6dB attenuation between the amplifier and speakers. You can't hear those quiet passages anymore, yet you know they are there because you just heard them before adding the attenuator. They don't add anything more to the music just because you turned up the volume control.

Of course in optics there are other factors affecting the information in the image, but the 2XTC is about as analogous to a linear amplifier as you can get.

[Lou Jost]

The information is not lost in downsizing, because the display and file generation are not redoing a Bayer interpolation. The original image only had one color actually measured at each pixel. The new image made four color measurements (R G G B) in the same space. When you downsize, the new RGB triplet of the downsized image contains real data for each color. You've replaced purely interpolated colors by some actual measured colors.

[ray_parkhurst]

The information is not lost in downsizing, because the display and file generation are not redoing a Bayer interpolation. The original image only had one color actually measured at each pixel. The new image made four color measurements (R G G B) in the same space. When you downsize, the new RGB triplet of the downsized image contains real data for each color. You've replaced purely interpolated colors by some actual measured colors.

The additional information captured due to the extra magnification of the 2XTC is indeed lost. What you are describing is information that is available to the sensor without the TC, but because the sensor is imperfect, it cannot properly capture that info.

[Justwalking]

I leave this article here for furter discussion

https://epxx.co/artigos/dof_en.html

[ray_parkhurst]

I leave this article here for furter discussion

https://epxx.co/artigos/dof_en.html

Hmm, a very quick look shows that article is full of misinformation. Might be worth a discussion on separate thread.

[Lou Jost]

What you are describing is information that is available to the sensor without the TC, but because the sensor is imperfect, it cannot properly capture that info.

Yes, agreed.

The additional information captured due to the extra magnification of the 2XTC is indeed lost.

But this is not true. By expanding the image over the sensor, you now have four pixels where there was once only one, so now you are measuring all three colors at every (original) pixel. That information is not lost when you downsize the image back to the original size; after downsizing, the R G and B values of the image at each pixel include actual measured information for each of those colors, not just extrapolated information from neighboring pixels (for two of the colors) as in the original image.

[ray_parkhurst]

The additional information captured due to the extra magnification of the 2XTC is indeed lost.

But this is not true. By expanding the image over the sensor, you now have four pixels where there was once only one, so now you are measuring all three colors at every (original) pixel. That information is not lost when you downsize the image back to the original size; after downsizing, the R G and B values of the image at each pixel include actual measured information for each of those colors, not just extrapolated information from neighboring pixels (for two of the colors) as in the original image.

You're still thinking of the information that would normally be contained in the non-TC image. I absolutely agree that this information is not lost, and when downsized by 2x, the resulting image is much closer to what an ideal sensor would have captured without the 2XTC.

But the image with the 2XTC contains additional information, at higher spatial frequencies, which the 2x downsized image simply cannot support. This is the "additional information captured due to the extra magnification of the 2XTC" I am referring to. I could have been clearer, but I don't consider information that should be captured without the TC to be part of this "additional information". The smallest subject features that are resolved by the 2x magnified image cannot be resolved when downsized 2x, and are lost.

edited to add:

Hmm, this brings up a question I keep forgetting to ask...

When downsizing, we essentially re-sample the image at a lower spatial frequency. The original image contains higher frequency information, and this information will alias back into the desired image data unless properly filtered. So, do image downsizing algos implement a digital filter to brick-wall this information? I assume so, but I have noticed that many downsized jpgs lack sharpness. I will often downsize separately before doing any sharpening, because even if I sharpen and downsize in the same step, the resulting image is often not as sharp as I expect it to be. Is this the result of aliasing? If anyone has a favorite online resource to answer these questions, it would be much appreciated.

[Lou Jost]

I am not sure I agree with the verbal distinctions, but I do agree that there is info in the TC image which is not contained in the downsized image.

I guess when you say "information that would normally be contained in the non-TC image", you mean "information that would have been contained in the non-TC image if every pixel measured complete information about all three colors". If that's what you mean, I agree.

But this means that information which is not present in the original captured image, but is present in the captured TC image, does get transferred to the downsized captured image.

[ray_parkhurst]

I guess when you say "information that would normally be contained in the non-TC image", you mean "information that would have been contained in the non-TC image if every pixel measured complete information about all three colors". If that's what you mean, I agree.

Yes, exactly.

But this means that information which is not present in the original captured image, but is present in the captured TC image, does get transferred to the downsized captured image.

Yes, information at low enough spatial frequency is not lost.

The sensor is not "terrible" at capturing the non-TC image, so I expect that the downsized image contains essentially all information in the aerial image that can be captured by a sensor of same size. So I have long considered 2x downsizing to provide a near-perfect image, within the scope of the applied image processing parameters.

[Justwalking]

I did experiment also. Both with same parameters to .jpg convertion

It is 100% crop with 1.5X industrial Moritex teleconverter without sharp or anything and resized by 1,5.



And without TC and resizing. It looks better as 100% crop