Fast coding unit division method based on CNN

In order to reduce the computational complexity of High Efficiency Video Coding(HEVC), this paper introduces a three-branch convolutional neural network(CNN) based on texture, which aims to predict the code unit(CU) partition mode to achieve fast segmentation. Firstly, a three-branch structure is used to adapt to the feature extraction and prediction of CUs in different depths. Secondly, the texture is enhanced by introducing Sobel filter and combining attention mechanism. Thirdly, by fusing the spatial feature information extracted by Haar wavelet decomposition, the ability of model feature extraction is enhanced. The experimental results show that compared with the original HM16.20 encoder, the proposed method reduces the intra coding time by 75.228% on the test sequence.