Fast Intra Mode Coding Based on Convolutional Neural Network
Chengsi Lin, Qingming Yi · Academic Journal of Engineering and Technology Science · 2020
In order to better adapt to the different texture features of video images, the number of intra coding modes in the new generation of video coding standard h.265/HEVC (high efficiency video coding) has increased to 35, which not only achieves better coding performance but also increases the computational complexity. In order to reduce the complexity of intra coding, a fast intra prediction method based on convolutional neural network (CNN) is proposed. For 4x4 or 8x8 PU, this paper gets the list of candidate modes by CNN, skipping the rough mode decision (RMD) process of prediction unit (PU). In this paper, the algorithm is embedded in HEVC coding framework, which effectively reduces the redundant intra prediction process in all intra configuration. The experimental results show that compared with HEVC official test model (HM16.12), the coding time of the algorithm proposed in this paper is reduced by 28.08% on average, while that of BD_BR and BD_PSNR is only 1.14% and - 0.055db.