A Fast Approach for Coding Tree Unit (CTU) Partitioning Using Edge Information in Video Compression
Md. Zahirul Islam, Tanvir Ahmed Redoy, Mir Safwan Marzouq, Rizoan Toufiq · 2023
The HEVC standard brings significant enhancements to bitrate compared to the previous H.264 standard. Nevertheless, it introduces the challenge of intricate encoding. A primary contributor to this complexity is the partitioning of coding tree units (CTUs) necessitating an exhaustive search for optimizing rate-distortion trade-offs. The quest for optimal partitions for each video frame presents a major source of complexity due to the extensive exploration of numerous combinations. This process entails evaluating various partition sizes and configurations to strike a balance between bitrate and quality for each coding unit (CU). The complexity escalates exponentially with the growing number of CUs and available partition options as the search space expands. To address this challenge, we propose a complexity reduction methodology for intra-mode HEVC. Rather than relying on rate-distortion optimization (RDO), our approach harnesses an edge detection model to anticipate CTU partitions. Empirical results validate the efficacy of our approach, yielding encoding time reductions of up to 74.60% and 72.98% using the Roberts and Sobel edge detection techniques respectively across 12 standard test video sequences. These encoding time reductions, however, entail a negligible Bjøntegaard delta bitrate increments of 4.60% and 3.93% for the Roberts and Sobel methods respectively relative to the HM reference software. Notably, our proposed technique surpasses other state-of-the-art methods in terms of time efficiency.