A study of The Statistics of Motion Vectors with Applications to Video Coding

• Main Authors: Yu-Sharng Weng, 翁裕祥
• Other Authors: Shen-Chuan Tai
• Format: Others
• Language: en_US
• Published: 1996
• Online Access: http://ndltd.ncl.edu.tw/handle/32315105178009105296

碩士 === 國立成功大學 === 電機工程研究所 === 84 === Motion estimation is one of the most effective techni- ques used in interframe predictive coding. In this Thesis, three fast block-matching algorithms for motion estimation are presented. All the three algorithms utilize the corre- lations among the motion vectors of adjacent blocks. A first-order Makov process is used to analyze the inter relations among of motion vectors of adjacent blocks both in spatial and temporal domains. Based on the analysis results, two algorithms are first proposed. The first algo- rithm uses the neighboring, both in temporal and spatial domain, motion vectors to predict the motion vector of the current block.In order to increase the accuracy of estima- tion and avoid trapping into a local minimum, multicandi- dates are adopted in this algorithm. The second algorithm use the search area that is the union of five lollipop- shaped regions constructed by a motion vector. The third algorithm combines the two algorithms. The qualities im- prove significantly but its computation complexity is greater than that of the two algorithms . Several video sequences for testing the three proposed algorithms are employed and the results are compared against the results with those by other fast block-matching algorithms.