Fast Coding Quad-Tree Decisions Using Prediction Residuals Statistics for High Efficiency Video Coding (HEVC)

Abstract

High Efficiency Video Coding (HEVC) is the latest video coding standard to meet market demands for real-time high quality video codecs. As compared to its predecessor H.264/AVC, HEVC can achieve significant compression gains but with higher encoding complexity. Therefore, for real-time applications, significant encoding time reduction is still necessary. In the HEVC test model, the large number of coding quad-tree decisions to be tested during rate-distortion optimization would result in high encoding time. Hence, we propose a method to reduce the high encoding time by pruning the coding quad-trees using prediction residuals statistics. Experimental results from HM16.3-based implementations show that the proposed residual-based pruning method can reduce encoding time by an average of about 44% with an average of about 1.0% coding loss.