Efficient screen content intra coding based on statistical learning

Screen content videos play an important role in recent popular mobile technologies and cloud applications. With the new coding tools (including intra block copy, palette mode etc.) adapted, high efficiency video coding (HEVC) based screen content coding (SCC) achieves high coding efficiency but requires a very high computational complexity. In this paper, we propose a fast intra coding algorithm for SCC based on statistical learning. First, we analyze the complexity distribution of SCC encoders. Then, a classifier is designed to determine whether current coding unit (CU) should be split into four sub-CUs without performing intra prediction procedure at current CU depth level or not. If current CU should not be split into sub-CUs, another classifier is used to decide either 35 traditional intra modes or SCC modes (i.e. intra block copy mode and palette mode) will be performed at current CU depth level. Three kinds of features are chosen to generate the classifiers: (1) coding information of current CU; (2) texture information within current CU; (3) information of nearby CUs. Experimental results show that the proposed algorithm achieves a computational complexity reduction with 45% on average, and only 1.6% BDBR increase compared to the original coding in SCC reference software.