A Study Of Fast Inter Prediction Algorithm Based On Video Statistical Properties For HEVC

With the growing demand for video quality, and HD and ultra-high definition video will become more and more popular, which brings great challenge to video compression. In 2013 January, the JCT-VT published a new generation of video coding standard HEVC. The standard has a great improvement compared to the previous generation H.264/AVC. In HEVC, the size of the maximum code block is increased from 16 × 16 to 64 × 64, and HEVC adopts more intra prediction and inter prediction modes, which makes the prediction more accurate. The compression ratio of HEVC can be improved over the previous generation standard about 50% under similar quality. HEVC standard can support the video resolution up to 8192×4320. It has become hot spot of the research in current video compression.Although the HEVC standard achieves higher coding efficiency, it scarifies morecomplexity. There are many researchers have done a lot of improvements in HEVCoptimization. However, these studies did not take into account the efficient utilization ofthe correlation of the inter-frame prediction to reduce the time complexity. In this paper,we present three fast methods in inter prediction:The first is rapid selection algorithm of CU depth and fast mode selection of inter-prediction based on combination of the correlation between adjacent frames. The algorithm makes full use of the correlation between adjacent frames in depth and prediction mode. It has achieved a decrease in the number of the divide times of CU and the number of prediction model. Secondly, a fast algorithm in the mode of Merge is proposed. The algorithm replaces complex RDO calculation with a simple cost calculation method. It has reduced the computational complexity. Finally, we have the proportion of skip model and asymmetric modes. Skip model is more accurate and needs less bit coding. Non-symmetric model accounts for only a small proportion of the optimal mode but high computational complexity. Fast skip mode algorithm is presented and the asymmetric prediction mode is canceled.Three optimization algorithms based on inter-prediction are presented in this paper. It can reduce the computational complexity effectively. In order to verify the efficiency of the proposed rapid method, we test a series of video. The average increment of bit stream is within 3%, the average decline in PSNR is within 0.1d B. The three rapid methods save processing time for 52%, 22% and 54% of the processing time of that in original HM10.0 algorithm. The experimental result shows our rapid algorithms can improve the coding speed significantly with acceptable image quality and compression efficiency.