Probability density estimation based on the parzen window for fast Cu size decision of HEVC intra coding

Abstract

The critical problem with the practical implementation of the high efficiency video coding (HEVC) encoder is the computational complexity. Complexity increases because of its novel techniques, such as quad-tree-based block partitioning. Thus, for years, researchers have used various methods and studied multiple algorithms to reduce the computing time of the HEVC encoder. However, a fast algorithm using the Parzen window has not yet been investigated. The Parzen window is a non-parametric method that estimates probability density using statistical samples and information. The proposed method is utilized to predetermine the size of a coding unit and statistical data for the Parzen window is updated periodically. Experimental results show that the proposed algorithm reduces encoding time by 51.96 % with a small Bjøntegaard-Delta bit-rate increase of 0.56 %. © Proceedings of 2018 the 8th International Workshop on Computer Science and Engineering.