Hybrid Intraprediction Based on Local and Nonlocal Correlations
Tao Zhang,Xiaopeng Fan,Debin Zhao,Ruiqin Xiong,Wen Gao
DOI: https://doi.org/10.1109/tmm.2017.2775223
IF: 7.3
2018-01-01
IEEE Transactions on Multimedia
Abstract:In the latest video coding standard, namely, high-efficiency video coding (HEVC), intra coding efficiency is significantly improved by a quadtree partition structure and more intra prediction modes. For intra coding, 35 intra modes are employed including 33 angular intra prediction (AIP) modes that are effective for blocks with strong directions, and a planar mode and a dc mode that are used to predict smooth regions. However, intra prediction in HEVC still cannot handle complicated blocks well. To deal with this problem, this paper proposes a hybrid intra prediction method to improve the intra prediction efficiency. The proposed hybrid infra prediction method consists of three parts: adaptive template matching prediction (ATMP) by exploring nonlocal correlation, combined local and nonlocal prediction by exploring both local correlation for AIP and nonlocal correlation for ATMP, and combined neighboring modes prediction that can generate smooth prediction by exploring more local correlations. Experimental results suggest that the proposed hybrid intra prediction method achieves 2.8% RD-rate reduction on average for luma compared to HEVC reference software HM-14.0 under all intra main configurations. The gain can be up to 6.9%. When integrated into joint exploration model -1.0, the proposed method still can achieve 1.2% RD-rate reduction for luma.