Fast Intra Coding Partition Decisions for Feature Video Coding

Future video coding (FVC) can support 4K high-resolution videos to replace the previous coding standard, high efficiency video coding (HEVC). In particular, FVC adopts quadtree with binary tree (QTBT) to improve coding efficiency; however, encoding time increases heavily. Thus, we propose fast partition algorithms for FVC intra coding. Fast partition algorithms include spatial correlation method and deep learning method. The spa-tial correlation method uses gradient variances of pixels to determine early skip and early termination for QT depths 0 and 1, respectively. The deep learning method uses convolu-tional neural networks (CNNs) to predict the QTBT coding structure at QT depth 2 with its corresponding coding trees. Experimental results show that the proposed method can reduce encoding time by 12.49% on average but increases Bjontegaard delta bit rate (BDBR) by about 0.54%.