In multimedia, the quality engineers are optimizing for is perceptual. Eyes, ears, and the brain processing their signals are enormously complex, and there’s no way to replicate everything computationally. There are no “objective” metrics to be had, just various proxies with difficult tradeoffs. Modifying video is particularly thorny, since like I’ve mentioned before on this blog there are various ways to subtly bias perception that are nonetheless undesirable, and are impossible to correct for....