Fast intra prediction based on the texture feature for H.264/AVC encoding

Intra prediction in H.264/AVC high profile supports rich block types and prediction modes to achieve high coding efficiency. In order to obtain the best block type and prediction mode, the full search algorithm requires a lot of calculations. This study aims to reduce the computational complexity of intra prediction by choosing both appropriate block types and a few candidate modes. In the proposed algorithm, each macroblock is divided into sixteen 4x4 sub blocks and edge strength and direction of each sub block are obtained based on the edge detection templates. Then the mean absolute deviation of edge strengths among sixteen sub blocks is used to measure the smoothness degree of the macroblock and select the appropriate block types. For the selected block types, the candidate modes can be further chosen according to the edge direction of the block. Experimental results show that the proposed fast algorithm can achieve about 74% computational time saving on average for encoding the all intra-frame sequence with negligible loss of coding performance compared with the original full search algorithm. © 2015 Academic Journals Inc.