HDR Video Coding Based on Perceptual Optimization

High dynamic range (HDR) video is a high-quality video format that improves contrast and colors. Encoding HDR directly faces issues like visual redundancy, uneven bitrates, and subpar subjective visual perception. As a solution, a perception-based optimization algorithm for HDR video encoding is proposed. Firstly, leveraging the human visual system's perceptual redundancy, a perceptual lossless preprocessing model using Karhunen-Loève (KL) transform and a subjective perceptual lossless threshold is created to cut visual redundancy. Subsequently, a coding tree unit (CTU)-level perceptual quantization model, based on human visual attention, is developed. This model estimates the perceptual saliency in spatial and temporal areas, directing CTU bitrate allocation for optimized bitrate that aligns with human perception. Experimental results show the algorithm effectively reduces visual redundancy and enhances coding performance. Compared to VTM 17.0, it reduces bitrate by an average of 22.67% (up to 53.03%), reduces coding time by 26.91% (up to 51.79%) and increases average PSNR by 5.07% (up to 10.46%).