H.264 visually lossless compressibility index: Psychophysics and algorithm design

Abstract

Although the term ‘visually lossless’ (VL) has been used liberally in the video compression literature, there does not seem to be a systematic evaluation of what it means for a video to be compressed visually lossless-ly. Here, we undertake a psychovisual study to infer the visually lossless threshold for H.264 compression of videos spanning a wide range of contents. Based on results from this study, we then propose a compressibility index which provides a measure of the appropriate bit-rate for VL H.264 compression of a video given texture (i.e., spatial activity) and motion (i.e., temporal activity) information. This compressibility index has been made available online at [1] in order to facilitate practical application of the research presented here and to further research in the area of VL compression.