Toward Generalized Psychovisual Preprocessing For Video Encoding

Deep perceptual preprocessing has recently emerged as a new way to enable further bitrate savings across several generations of video encoders without breaking standards or requiring any changes in client devices. In this article, we lay the foundation for generalized psychovisual framework encoding and describe one its promising instantiations that is practically deployable video-on-demand, live, gaming, user-generated content (UGC). Results using state-of-the-art advanced coding (AVC), high efficiency (HEVC), versatile (VVC) show average [Bjontegaard delta-rate (BD-rate)] gains 11%–17% are obtained over three reference-based quality metrics [Netflix multi-method assessment fusion (VMAF), structural similarity index (SSIM), Apple tool (AVQT)], well proposed nonreference International Telecommunication Union-Telecommunication?(ITU-T) P.1204 metric. The on CPU shown be twice faster than $\times 264$ medium-preset encoding. On GPU hardware, our approach achieves 714 frames/sec 1080p (below 2 ms/frame), thereby enabling use very-low-latency live game streaming applications.