Piecewise Linear Model Based Local Illumination Compensation Inter Prediction for Video Coding
Chen Wang,Xin Jin,Huanbang Chen,Haitao Yang,Kedeng Tong,Yuqing Yang
DOI: https://doi.org/10.1109/pcs56426.2022.10017996
2022-01-01
Abstract:In video coding, the illumination information of video scenes is usually hard to be compressed due to the complex and unpredictable illumination variations. To simplify the problem, many prediction algorithms assume a linear correlation of illumination variations existing between frames by constructing corresponding linear model (LM), such as Weighted Prediction (WP) method. The assumption is suitable for uniform illumination variations with lager area, but ineffective in sharp illumination variations in small area. In this paper, we propose a piecewise linear model based local illumination compensation (PLMLIC) approach to further compensate sharp illumination variations in small area. When PLMLIC is applied to a coding unit (CU), we first use multiple reference lines, i.e. allow not only the nearest reference line but also long-distance reference lines to be the neighbouring samples of the current CU. Then the neighbouring samples and their corresponding reference samples are classified into 2 groups, based on which PLMLIC parameters are derived for each group. Finally, the current CU is predicted by the corresponding PLMLIC parameters. Experimental results show that 0.23% BD-rate savings on average can be achieved for lowdelay configuration based on Enhanced Compression Model (ECM) beyond VVC.