Font Size: a A A

Research And Parallel Implementation Of Motion Estimation Used In Video Encoding

Posted on:2012-11-29Degree:MasterType:Thesis
Country:ChinaCandidate:C ZhuFull Text:PDF
GTID:2268330425990440Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
The motion estimation algorithm, which is the primary technology in achieving the video high compression rate, is a very high computational complexity, so its application in the area of real-time video coding is seriously limited. On the other hand, with the expansion of high quality video application in the video telephone, video conference, digital surveillance, and high definition television and other scopes, the research of fast motion estimation algorithm has important practical value and broad prospect of application.The domestic and foreign studies show that there are many research results on fast motion estimation algorithm. The major technology includes initial search point prediction technology, terminating threshold technology, and reducing unnecessary reference frame technology and so on. The use of these technologies is effective to reduce the computational complexity of motion estimation algorithm. Currently, a lot of fast motion estimation algorithms have been proposed, but there still is much space for further optimization on motion estimation algorithm.According to the insufficiency of the UMHexagonS algorithm used in the X264, the improvements of three aspects are proposed in this paper, which include duplicate detection about origin, changing the motion search strategy and dynamic adjustment of search window size. The test results show that encoding speed has improved8.223%~16.063%after the improvement with basically remaining unchanged in the image objective quality and encoding rate.Since seven mode motion estimation needs to be run in turn when a macro block in image regional in H.264/AVC standard is encoded, and according to the X264encoder’s characteristics in implementing the mode selection, a program frame executed in parallel in mode selection is proposed in this paper, and is respectively implemented using OpenMP technology and the Windows multi-threads technology. Test results show that compared to the serial structure, the parallel structure designed obtains14.208%~40.984%enhancement in encoding speed, but no obvious influence on the image objective quality and code rate. When multimedia instruction set for acceleration included in the X264encoder is closed, encoding speed will increase32.757%~53.089%.
Keywords/Search Tags:Video coding, Motion estimation, Parallel computing, Multi-threads
PDF Full Text Request
Related items