Research And Design Of H.264 Inter-frame Prediction Algorithm

Inter-frame prediction as one of the key technologies of H.264 video coding standard can significantly reduce the temporal redundancy and improve the ratio of compression. However, the high computational complexity of inter-frame prediction algorithm greatly affects the application of H.264 video coding standard, especially in the field of real-time video coding. In order to solve the problem of great high computational complexity, improve the video encoding performance, and shorten the coding time, the classical motion estimation algorithm of UMHexagonS in inter-frame prediction is researched on.The UMHexagons algorithm has been deeply analysised and researched on to improve the coding performance of inter-frame prediction algorithm for motion estimation. A grand cross prediction search algorithm is proposed to reduce the redundancy search in UMHexagonS algorithm. The grand cross prediction seach algorithm can predict the possible distribution area of the best point, which narrow the search range, and then uses corresponding search template in the selected area to search the best point. Experimental results in JM model show that the grand cross prediction search algorithm compared to the UMHexagons algorithm in strenuous exercise scenes can decrease motion estimation time of inter-frame prediction by 30%-50%, and keep similar performance in micro motion scene. To further improve the grand cross prediction search algorithm performance in micro motion scene, the moving scene classification of inter-frame prediction algorithm is proposed. The moving scene classification algorithm of inter frame classify the moving scene as micro motion scene and strenuous exercise scene. The micro motion scene uses with local arear diamond search algorithm to narrow the search range and decrease the number of search points, and the strenous exercise scene uses with grand cross prediction search algorithm to predict the area where the best point will be. Through the experiment results in JM model, the time of moving estimation in the moving scene classification algorithm is reduced by 20%-30% in micro motion scene, and is reduced by 40%-50% in strenous exercise scene compared to the time in the UMHexagonS algorithm. Due to the difference of PSNR is within 0.02 dB to UMHexagonS algorithm, the images keep the similar quality.To test the performance of the grand cross prediction search algorithm and moving scene classification search algorithm in embedded system, the verification system based on OMAP3530, which contains hardware and software system and embedded Linux operating system platform to build.The test results of the verification system prove that from the aspect of reducing frame encoding time the grand cross prediction search algorithm keeps better realization in strenuous exercise scenes, and the moving scene classification algorithm can get better realization of in both strenouse exercise scenes and micron motion scene.