A weighted intra prediction based on statistics and classification for HEVC

摘要

I frame or I slice which adopts the intra prediction is a key part of video coding. Intra prediction is important because the prediction accuracy directly affects the efficiency of the following transformation, quantization and entropy coding. According to the case of intra mode selection in HEVC, it is necessary to improve the prediction for the two most frequently used modes, DC and PLANAR. A significant approach to improve prediction is the adaptive classification of pixels and determination of weights. In this paper, we proposed a system to train weights based on the mathematical statistics and classification with the least-squares method to obtain appropriate prediction weights for these two modes. Compared to the HEVC reference software HM8.0, the proposed algorithm can reduce average 0.24% bit rates for videos with more details.