Key Algorithms Research In Image And Video Compression

The era of digitalization and information technology have brought conviniences to our everyday life. Multimedia technology applications are much more widely. On the other hand, however, the exploding size of digial image and video has brought large burden to storage and transmission. Data compression is able to remove redundant information from the large amount of data and retains only small amount of independent necessary information. The video coding technology has been widely used in a lot of areas. With the increasing application of low bitrate video, especially video on network, wireless network and video conferencing, video coding standards such as H. 264/AVC have more and more mature and complete. Research in video coding has draw more and more attention.Video coding based on Discrete Cosine Transform (DCT) is main stream of video and image compression standards. The precision in fixed-point computation of DCT directly influences the quality of image in video and image compression. So the precision of DCT has gained much attention. In the other conditions remain the same circumstances, with more decimal digits, said part of the algorithm is also a higher accuracy.In this thesis, we analyzed the dynamic range of 2D DCT and IDCT and proposed the optimal word length for DCT and IDCT. Compared with existing research, we have 3 extra bits at least for representing fractional parts of data. According to detailed analysis of DCT and IDCT, At most 7 extra bits can be used for representing the fractional parts of data elements.In fractional pixel motion estimation in H. 264/AVC, Sum of Absolute Transform Difference (SATD) using Hadamard Transform is able to improve the compression performance. However, SATD is less efficient in computation compared with sum of absolute difference (SAD). Existing algorithms mainly concerns the suboptimal fast search based on shapes to have a trade-off between computational complexity and compression performance. In this thesis, we expose the computational redundancy in the SATD computation for fractional-pel motion estimation in H. 264/AVC. A SATD-Pair based algorithm for half-pixel motion estimationis proposed in this thesis. The proposed algorithm does no harm to estimation accuracy and rate-distortion curve.