When you establish a 50/50 mix of any original digital signal with a negative image of the same signal after passing though the encoding process, the resulting visual differences are the errors generated by the encoding.
source: Avid DNxHD white paper (in our wiki), p.7
This struck me as a brilliant idea (I don’t know shit about editing). I did this to compare the effects of different encoding parameters with ffmpeg. I used ffmpeg itself to extract the same frame from the original and from different encoded versions. Then in Gimp I opened those stills as separate layers. I placed the original version at the top, applied “Linear Invert” and reduced opacity to 50% to it. Then I compared the artefacts visible in the resulting mix with each underlying layer.
edit: more at