Rotational Weighted Averaged Template Matching for Intra Prediction

Abstract

Intra prediction is an important step in video coding, which improves compression efficiency by reducing spatial redundancy. High Efficiency Video Coding (HEVC) has a total of 35 prediction modes, including 33 angular modes and a DC mode as well as a Planar mode. These modes usually exploit local directional signal correlations. In order to represent prediction blocks with non-local complex texture and content, template matching was proposed. However, there may be blocks exist in an image with the same content but with different rotation angles. In this paper, we propose an improved intra template matching prediction method, which extends the intra template matching technique by making the matched template rotated during template matching process. Also, in order to improve the efficiency of prediction, this paper adopts a weighted template matching method to predict the target block by averaging sum of the weighted corresponding remote blocks of three most similar remote templates. The experimental results show that for All-Intra configuration, the overall coding gain in BD-rate is about 5.1% at most and 0.63% on average comparing to averaged template matching in HEVC. Moreover, our approach can achieve 13.2% coding gain at most and 2.0% on average comparing to HEVC test model HM16.9.