SSIM o similar for Reflex photo Camera?

[Gannjunior]

Problem: i would like to support my eyes in the attempt to be sure that a full frame camera as Canon 5d will give a better output than an APS-C ( 1.6x crop factor) camera as Canon 40d.

Even if there are differents factors (subjective and objective) that influence our eyes perception - same thing in our video test - I was thinking if there are some ways, similar to SSIM or PSNR, that i could apply in photography.

In our video test we apply our .dll SSIM in one images sequence. In the photography you should apply the same reasoning only in one image, in one frame. ok?
My question: can i "simply" apply SSIM only on 1 frame (that is my image) or you know the existence of some specific tool dedicated to the photography that can evaluates the luminance difference (that is what our eyes mainly can perceive) between two similar shots?

another thing, that i believe: in any way our comparison will return hard. When we compare a video sequence, for example to evaluate the efficiency of one codec against another one, we are able to start from the SAME (for example DVD) source.
But in my case it is impossibile to get exactly 2 IDENTICAL shots from two differents camera. (you can equal the exposure and the aperture but not you can't get the same identical framing...).

Thanks in advance.

ciao!

[Soulhunter]

To compare the output via some metric, you'd need exactly the same input!

Seems kinda impossible to me... :\


Bye

[Sagekilla]

What you basically want to do is, given today's commonly available technology, impossible. As you said, it's near impossible to get identical framing for the same pictures. Also, you have to deal with different resolutions. Do you resize downwards for the bigger image? Upwards for the smaller image? What about the actual frame size; Are they the same aspect ratio? There's a ton of different factors why it wouldn't work.

Sorry, SSIM and PSNR is mostly meant for comparing a source followed by a compressed image in this context, not for comparing two different cameras. You'd also need a master image, a "perfect" image to compare the two cameras to.

[Tim Bate]

I agree with both of the above posts, but my guess is you still need to try something!

There are other methods rather than AVISynth that I might have a look at for this. Superresolution is a method that is faced with similar problems; you have to get two shots from two different positions perfectly lined up. You might do well to take a browse for papers on Superresolution or just image registration. If you save the video as sequences of frames, and then apply image registration to each frame, that should go some way to mitigating the problem of framing.

As for having the perfect referance, I am guessing firstly that you are shooting from a tripod, and shooting a still scene. Movement is going to make things REALLY hard! If you are shooting from a tripod, I suggest using footage from your best camera and then frame stacking (averaging) to eliminate as much noise as possible. This will leave you with any systematic noise, but eliminate any random noise. To reduce systematic noise, you will have to average with another camera, and life suddenly becomes hard again.

Just my two cents.
Cheers!

[Sagekilla]

My 2 cents on the matter: When it comes to digital photography, measuring using metrics is insanely hard. Not to mention it doesn't matter 99% of the time if you're very good with your camera and work past it's limitations.