Adaptive weighted prediction based on moving area extraction and local brightness variation detection

In this paper, an adaptive weighted prediction is proposed to improve the coding efficiency. Conventional weighted prediction methods are optimized for specific sequences with global brightness variations (GBVs) such as fade-in and fade-out. However, if there is uncovered background by motion between current picture and reference picture, weighted prediction parameter (WPP) could not be derived accurately. And if there are local brightness variations (LBVs) between the pictures, it is not efficient to derive WPP over the entire picture. In order to solve above-mentioned problems, two kinds of technologies are added on conventional weighted prediction. First, moving area extraction (MAE) technology is devised to classify a picture into moving area and background area, and then two sets of WPP are derived for each area. Secondly, local brightness variation detection (LBVD) technology is added to detect the area including LBVs, and then the third set of WPP is derived for the detected LBV area. The proposed scheme is implemented on the HEVC Reference Software HM 10.0, and shows that maximum coding efficiency gain is up to 11.0% in luminance.