| Summary: | 碩士 === 國立海洋大學 === 電機工程學系 === 90 === In modern video coding, motion estimation plays an important role because of its ability of high compression efficiency. However, motion estimation generally involves heavy computation load, and many fast search algorithms were proposed to alleviate this problem. Despite these efforts, some important issues deserve further study. First, nearly all search algorithms for motion estimation, e.g. Diamond Search [8] and Cellular Search [9], only consider the aspect of the search pattern, and finding the optimal search pattern has been the central issue. Moreover, these methods often employ fixed search routine, which inevitably incur unnecessary computations, particularly in slow image video sequences. Aiming to solve these problems, this thesis presents two estimation modules. Using the variation information of the processing block, the proposed modules are capable of choosing the pattern most suitable for the processing block, and obtaining a better initial search center for the current search window. Experimental results are provided to show the effectiveness of the proposed modules, and their flexibility when incorporated into other search algorithms.
|
|---|
