| In visual IP telephony's terminal systems, as the image for video information needs a mount of bits, high quality of channel bandwidth and large storage space are required, so video compression has been the effective ways which to solve this question. Motion estimate is the key skill in video compression coding, a way of between frames coding. It bases on the theory of temporal correlation between frames, to reduce temporal redundancy. An outstanding method of motion estimate is how to get an accurate motion vector. Finding out the most resembled part as the matching part in the neighbor by using the present part, pointing at it by motion vector, then finding out the different part between the matching part and the present part, at last rebuilding the compressed frame by the matching part and the different part. Therefore, in the video compression coding whose key skill is motion estimate, the speed of searching motion vectors and the accuracy of motion vectors are the most important factors to video effect. This thesis focuses on researching the classical algorithms of motion estimate, analying the advantages and disadvantages of exhaustive search algorithm, three-step search algorithm and diamond search algorithm, providing the scientific theory and experiment data. Basing on the relationship between the quality of video and the running time of algorithm and the motion vector of space-time correlation and center biased distribution, put forward predict three-step search algorithm, predict diamond search algorithm and adaptive diamond search algorithm, testify them on the experiment. The results indicate that the new algorithms are better than the former algorithms in searching time and compressing quality. Finally we design a scheme of real-time video compression system on the DSP chip and analyze them.
|
PDF Full Text Request