On the Effectiveness of Video Recolouring as an Uplink-model Video Coding Technique

Abstract

For decades, conventional video compression formats have advanced via incremental improvements with each subsequent standard achieving better rate-distortion (RD) efficiency at the cost of increased encoder complexity compared to its predecessors. Design efforts have been driven by common multi-media use cases such as video-on-demand, teleconferencing, and video streaming, where the most important requirements are low bandwidth and low video playback latency. Meeting these requirements involves the use of computa- tionally expensive block-matching algorithms which produce excellent compression rates and quick decoding times. However, emerging use cases such as Wireless Video Sensor Networks, remote surveillance, and mobile video present new technical challenges in video compression. In these scenarios, the video capture and encoding devices are often power-constrained and have limited computational resources available, while the decoder devices have abundant resources and access to a dedicated power source. To address these use cases, codecs must be power-aware and offer a reasonable trade-off between video quality, bitrate, and encoder complexity. Balancing these constraints requires a complete rethinking of video compression technology. The uplink video-coding model represents a new paradigm to address these low-power use cases, providing the ability to redistribute computational complexity by offloading the motion estimation and compensation steps from encoder to decoder. Distributed Video Coding (DVC) follows this uplink model of video codec design, and maintains high quality video reconstruction through innovative channel coding techniques. The field of DVC is still early in its development, with many open problems waiting to be solved, and no defined video compression or distribution standards. Due to the experimental nature of the field, most DVC codec to date have focused on encoding and decoding the Luma plane only, which produce grayscale reconstructed videos. In this thesis, a technique called “video recolouring” is examined as an alternative to DVC. Video recolour- ing exploits the temporal redundancies between colour planes, reducing video bitrate by removing Chroma information from specific frames and then recolouring them at the decoder. A novel video recolouring algorithm called Motion-Compensated Recolouring (MCR) is proposed, which uses block motion estimation and bi-directional weighted motion-compensation to reconstruct Chroma planes at the decoder. MCR is used to enhance a conventional base-layer codec, and shown to reduce bitrate by up to 16% with only a slight decrease in objective quality. MCR also outperforms other video recolouring algorithms in terms of objective video quality, demonstrating up to 2 dB PSNR improvement in some cases.