Improved Fast Mode Decision Algorithms For Inter-prediction Research Based On HEVC

This project is based on High Efficiency Video Coding,which is the latest video coding standard.HEVC has better compression performance relative to previous standards—about a half bit-rate reduction for equal perceptual video quality.To achieve such performance,it introduces more flexible coding structure and more precise mode decision strategy at the cost of complexity.Screen content video,are captures of typical computer screens or video with text overlay which has broad application.With abrupt scene changes,sharp edges and other special features,screen content video is getting more and more attentions,which makes research on it become meaningful.In HEVC coding process,reference picture management plays an important role in increasing video quality and coding efficiency.Algorithms have been proposed to optimize reference picture management but few of them are targeting at screen content video.The targets of this project are designing fast algorithms for inter-prediction.I proposed two improved mode decision strategies based on the original algorithms.Proposed algorithms take use of temporal and spatial correlations among pictures to reduce the processes that have negligible contributions to compression performance but take up a lot of time.At the stage of prediction mode selection,I propose an early termination algorithm based on temporal correlations between coding unit size and prediction mode.At the stage of motion estimation,I propose a motion vector merge algorithm based on spatial correlations of motion vectors for different size of prediction unit on motion-homogeneous regions.To guarantee the rate distortion performance,I set rational restrictions for these two strategies by reiterative simulation and analysis,which are rate distortion threshold and motion vector difference threshold.The combined algorithms are then implemented in the 15th version of HEVC test model.Experimental results demonstrate that compared to the state-of-the-art published method,the proposed algorithm can achieve about 23%encoding time on average saving while the rate distortion performance degradation is negligible.This project,by finding out most efficient coding structure and using scene change frame as longterm reference picture,proposes a new efficient algorithm on reference picture management for screen content video.An efficient coding structure could defines reference pictures and the number of them.Long-term reference pictures provide useful information for pictures after scene change.Experiments have been conducted to prove the effectiveness of proposed algorithm.The results show that efficient coding structure achieves 0.69%bits saving and 12%time saving without causing much degradation on video quality compared to original coding structure.Long-term reference picture algorithm achieves up to 7.2%bitrates saving while keeping video quality undamaged.