H. 264 Fast Frame Prediction And Interframe Motion Estimation Research And Optimization Of The Algorithm

Intra prediction can significantly reduce the spatial redundancy in the video image,and inter prediction can significantly reduce the temporal redundancy among videoframes, so these two technologies are very important in the H.264/AVC video codingstandard. This dissertation focuses on the fast mode decision algorithm of intraprediction and the motion estimation algorithm of inter prediction. In order to improvevideo coding performance, a fast mode selection algorithm for intra prediction whichbased on adjacent pixels’ differences is proposed, while for the motion estimation ofinter prediction, the existing classical algorithm——UMHexagons is optimized andimproved.As to the study of intra prediction algorithm, firstly, the detailed implementionprocess of the full search algorithm and the three-step search algorithm are studieddeeply. After analyzing the reason of huge computational complexity in these twoalgorithms, a fast mode selection algorithm for the intra prediction which based onadjacent pixels’ differences is proposed. This algorithm can quickly select the best intraprediction mode by computing the differences between adjacent pixels. Then thealgorithm is implemented in the JM platform and compared with the full searchalgorithm and the three-step search algorithm to verify its superiority. Finally, theproposed algorithm is improved by the existing algorithm MAD which based onmacro-block and then verified in the JM platform again. Experimental results show thatthe proposed algorithm can shorten the encoding time and improve the efficiency ofvideo compression compared with the full search algorithm and the three-step searchalgorithm.As to the study of motion estimation of inter prediction, firstly, the detailedimplemention process of several existing fast motion estimation algorithms are studied.After analyzing the advantages and disadvantages of each algorithm, the dissertationdetails the classic algorithm UMHexagons. Three aspects of the UMHexagons areoptimized and improved: add early termination conditions for the motion vector search,add the mechanism of dynamic adaptive adjustment of search window size and reduce the number of points in the square and hexagonal templates used in the UMhexagons。Finally, the optimized and improved UMHexagons is implemented in the JM platform。In order to verify the versatility of the optimized algorithm, a number of standard videoswhich have different movement types are chosen for testing. By analyzing the objectivedata and subjective images, we conclude that the optimized algorithm can significantlyreduce the motion estimation time and raise the encoding speed while maintainingalmost the same video images quality as the original algorithm, and also contribute tothe practical application of H.264/AVC.