Perceiving distortions in visual signals

This paper and the associated talk are largely an opinion piece on current research on visual quality assessment: where it is at, and where it is going. The paper and talk follow three themes. First I´ll discuss full-reference (FR) visual quality assessment (QA), and will opine that future gains on this largely-solved problem are likely to be found in modeling human behavior, which is a hard problem quite different from perceptual modeling. Then I will discuss the no-reference (NR) or blind QA problem, where satisfying results are now being obtained on still images. However, the blind video QA problem remains far from solution owing to a lack of regular statistical video models, among other things. Lastly, I will discuss the exceedingly difficult stereo (3D) visual QA problem. It is my hope that the talk and this paper will stimulate promising avenues of research.