Tree structured hybrid intra prediction

This paper proposes a new method to generate intra prediction images in intra coding, tree structured hybrid intra prediction. When generating intra prediction images, the proposed method selects the most suitable one from intra prediction images generated using two methods, pixel-based and block-based intra prediction. This selection is performed on macroblock or sub macroblock level based on a rate distortion optimization scheme. In addition, information indicating which prediction method is selected is described in the bitstream using a tree structure. The proposed method is initially described, and then the results of evaluation experiments are shown by comparison to conventional methods. Evaluation is carried out by implementing the proposed method in an H.264-based intra picture codec. The results of evaluation confirm that the proposed method improves PSNR by up to nearly 0.7 dB compared to conventional methods over a wide range of bit rates.