LOCAL-AWARE INTRA TEMPLATE MATCHING PREDICTION

Intra template matching prediction (IntraTMP) has been adopted in enhanced compression model (ECM) to code both screen and camera-captured content. In IntraTMP, a prediction block is copied from a reconstructed region of the current picture, wherein the location is derived using template matching. However, template matching may not find a perfect prediction block when the correlation between the template and the current block is limited. In this paper, a method of local-aware intra template matching prediction (LA-IntraTMP) is proposed, in which two prediction blocks are generated by IntraTMP and angular intra prediction, respectively, which are then fused together to obtain the final prediction block. Specifically, the template cost calculation is modified accordingly to construct an IntraTMP candidate list. Furthermore, the IntraTMP candidate list can be reordered based on template costs. Finally, the best IntraTMP candidate determined by the encoder can obtain a prediction block, fused with the angular intra prediction. Experimental results demonstrate that LA-IntraTMP can achieve 0.19% Bjontegaard Delta rate (BD-rate) saving compared to ECM-7.0 under all intra (AI) configuration.