Video-codec rater by image comparison algorithm?

Question

Hi,

perhaps anyone knows if this is possible. comparing image quality is almost imposible to describe without subjective influences. When someone rates an image quality as good there is at least one person, that doesn't think so. human preferences are always different.

So, I would like to know if there is away to "rate" the image quality by an algorithm that compares the original image to the produced one in following issues

• colour change(difference pixel by pixel
• blur rate
• artifacts and macroblocking

the first one would be the easiest one because you could check just the diffeence in colours and can give 3 values in +- of each hex-value both last once I don't know if this is possible, but the blocking could be detected by edge-finding.

and the king's quest would be to do that for more then just one image, because video is done with several frames.

perhaps you expert programmers could tell me, if such an automated algo can be done to bring some objective measurement divice into rating image quality. this could perhaps calm down some h.264 is better than x264 and better than vp8 and blaaah people :)

Andreas

1st posted here http://www.hdtvtotal.com/index.php?name=PNphpBB2&file=viewtopic&p=9705

Answer 1

This is a very complex topic with no clear solution. At the moment, standards testing is done with PSNR and subjective tests (like MOS). PSNR is by no means representative of subjective quality, but it's used everywhere. SSIM is supposed to be closer to subjective tests than PSNR, but it's still not as popular as PSNR.

If you're really interested in image metrics (still there's no clear choice for video metrics --- most of the time people average PSNR over frames) you can check out JND, in addition to SSIM and PSNR and this document has a few more including the ones I've mentioned.

Answer 2

It seems to me that any worth-while automated metric of image quality would be incorporated into the next revisions of compression algorithms, leaving you back where you were.