Low Complexity Encoding Of HEVC

With the continuous improvement of video quality, the intrinsic defects of the state-of-the-art H.264/MPEG-4 AVC video encoding standard are increasingly obvious. In response to the problem of video encoding and decoding as well as video transmission resulted from the increase of video resolution and video definition, the Joint Collaborative Team on Video Coding proposed a more powerful solution based on H.264/MPEG-4 AVC, that is High Efficiency Video Coding(HEVC) Standard. The new techniques adopted by HEVC increase the encoding time dramatically. In order to make HEVC be practical as soon as possible, many scholars have researched fast algorithm in intra and inter encoding.At present, many researchers utilize texture direction and texture complexity to save encoding time, and study how to predict the encoding depth of Coding Unit(CU) from the machine learnig perspective. This paper researches fast video encoding solutions from three parts, on the basis of the new techniques of HEVC.The first, this paper proposes a texture quadtree algorithm, which utilizes texture to build a texture quadtree. Experimental results show that the proposed algorithm saves 28.70% encoding time on average, while it only leads to 0.50% incresement and 0.03 dB degradation with respect to Bj?ntegaard Delta Bit Rate(BDBR) and Bj?ntegaard Delta Peak Signal-to-Noise Rate(BDPSNR) on average. At the same time, so as to reduce accumulated error, this paper proposes an algorithm, which is based on group of large CU. Experimental results illustrates that the proposed algorithm saves 31.14% encoding time on average, while it only leads to 0.24% incresement and 0.01 dB degradation corresponding to BDBR and BDPSNR on average.The second, this paper takes account of different growth of improper prediction and makes use of the Minimum Risk Bayesian Decision Rule to prediction the encoding depth of CU. Experimental results show that the proposed algorithm saves 31.71% encoding time on average, while it only leads to 0.86% increasement and 0.03 dB degradation with respect to BDBR and BDPSNR on average.The third, due to the frequently use of random access configuration in practical, to study inter frame encoding algorithm is especially significant. At first, this paper explores the CU structure character of stationary area. This paper proposes stationary area detection algorithm. Experimental results show that the proposed algorithm saves 19.14% encoding time on average, while it only leads to 0.04% increasement of BDBR on average. Then, this paper applies fast inter frame prediction mode algorithm to stationary area. Experimental results show that the proposed algorithm saves 31.47% encoding time on average, while it only leads to 0.44% increasement of BDBR.