Gradient based fast mode and depth decision for high efficiency intra frame video coding

Intra frame coding plays an important role in video coding. To get better performance, high efficiency coding standards exploit new techniques to improve the performance of intra-coding. A flexible partition structure and multiple prediction modes are applied to achieve accurate intra prediction. Different block sizes and prediction modes are traversed to find the best coding unit and accurate prediction direction. All these processes increase lots of computational complexity. To optimize the intra coding process, a gradient based algorithm is proposed in this paper to make fast mode and depth decision. A Sobel operator is applied to get the gradient information of pixels. Its statistical properties are studied to find the most probable prediction mode for further selection. Furthermore, the texture information guides coding unit partition to avoid unnecessary calculation. In order to reduce the encoding complexity, we started from the smallest coding unit and build a bottom-up partition process. Experiments are implemented on a high efficiency coding standard AVS2, and about 48% encoding time is reduced on average with negligible coding performance loss.