CNN oriented fast PU mode decision for HEVC hardwired intra encoder
Nan Song, Zhenyu Liu, Xiangyang Ji, Dongsheng Wang · 2017
The number of intra prediction modes in High Efficiency Video Coding (HEVC) has been increased up to 35. To the end of alleviating the complexity of intra coding, we bring in the convolution neural network (CNN) to obtain the candidate modes of current PU and adopt the corner detection algorithm to further reduce the candidate modes. The virtues of proposed algorithm include: Firstly, our algorithm skip the rough PU mode decision (RMD) process and get the candidate list from CNN directly. In other words, the computations are relaxed in our algorithm. Secondly, the inputs of CNN in proposed algorithm merely contain the source image pixels and quantization parameter (QP), this feature makes it friendly to high parallel hardwired encoder. As compared with HM-15.0, experiments show that our algorithm decreases the intra coding time by 27.92% while the corresponding BDBR augment is 1.15%. At last but not least, our algorithm possesses a stable coding performance. In specific, for the most sensitive sequence (Class F), our algorithm could save 27.10% intra coding time with 2.01% BDBR increase.