Enhanced intra coding with joint line and pixel prediction

Abstract

Existing video coding standards always employ line-based intra prediction for intra frame coding. However, it sometimes does not perform well for the bottom-right boundary pixels due to the low correlation with the referenced pixels. In this paper, we propose a novel joint line and pixel prediction (JLPP) method to compensate the performance degradation of boundary pixels. In JLPP, each block is partitioned into two subblocks, including a P-subblock covering bottom and right pixels and a L-subblock comprising the rest pixels. L-subblock will firstly be predicted and coded in a similar way to H.264/AVC intra coding. Secondly, a gradient-adjusted prediction scheme is applied to P-subblock with the aid of the reconstructed pixels in L-subblock. Therefore, the reconstruction qualities of the boundary pixels used to predict the subsequent blocks can be enhanced. Experimental results demonstrate that the proposed method achieves a PSNR gain of up to 0.63dB compared with H.264/AVC intra coding.