Tests that compare w/ source.

[KitKat]

Hi all,

I'm wondering if codec tests have been done by substracting the output frame from the original frame. That way you could actually "see" how much image data is thrown away. Wouldn't this be useful?

I ask because when I see comparisons between output frames of codec X and Y, it's not always obvious to me which is better, although it seems obvious for others.

[DigitAl56K]

Data is not necessarily "thrown away", but because of the nature of the algorithms typically used by lossy compressors there is a limit to their accuracy, and different algorithms will introduce certain artifacts.

There are various metrics for measuring error between source and encoded video. SAD, PSNR, and SSIM are a few common metrics.

If you want to see a difference display, try the encoder feedback window in DivX Pro, which allows you to view the difference in realtime during encoding, and actually amplify it for visibility.

[qwerpoi]

Well, if you are familiar with Avisynth, you can easily make subtract comparisons. For example,

a = AviSource("C:\uncompressed.avi") # the original file
b = AviSource("C:\divx.avi") # the compressed file
Subtract(a,b)

You can even download images from codec comparisons on the web and do the same thing. If you are not familiar with avisynth, check out the avisynth forums here or http://www.avisynth.org/. Of course, you have to be careful - subtract will indeed let you see any minute differences between two files, but that can be misleading. It can be useful, just don't make it the only basis of judgement.