Fast Algorithm Optimization For HEVC Inter Mode Decision

The High Efficiency Video Coding (HEVC) standard is the latest video coding standard. The main goal of the HEVC standardization effort is to enable significantly improved compression performance relative to existing standards in the range of 50% bit-rate reduction for equal perceptual video quality, and the cost is much more computational complexity than previous standard.This thesis does detailed research on the basic principle and key technology of HEVC, particularly in two parts of rate distortion optimization process in HEVC. One is prediction unit division of inter prediction, and another is reference frame selection. After Lots of experiments, reviewing existing research on these two points, this thesis make improvement and innovation based on the above, and design fast algorithms.First, in prediction unit division of inter prediction, a very strong correlation between current CU and the PU split mode in the corresponding position in parent CU was found via experiments and analysis. At the same time the current PU split mode also has a strong correlation with the corresponding position of adjacent frame. So an algorithm is designed using these two correlation, to avoid the exhaustive traversal by establishing prediction mode candidate. The results show that in Low-Delay scenario, this algorithm reduces encoding time by 50.10% at the cost of increasing coding rate by 1.22% and decreasing Peak Signal to Noise Ratio (PSNR) by 0.09% compared to the original algorithm.Second, the correlation of reference selection of Prediction Unit (PU) between two adjacent frames is analyzed. The redundant references are skipped for the current PU to reduce the coding computational complexity. The experiment results show that the proposed method saves 40.8% of encoding time at the cost of 1.42% of bit rate increase and 013% of PSNR decrease compared with the origin algorithm in HEVC in the Low-Delay scenario. And in Random Access scenario, saves 24.61% of encoding time at the cost of 0.42% of bit rate increase and 0.05% of PSNR decrease.