The Interpolation Filter Design for High Precision Motion Estimation

• Main Authors: Hung-Ran Yen, 顏鴻任
• Other Authors: Chung-Ming Kuo
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/57029839965568014607

碩士 === 義守大學 === 資訊工程學系碩士班 === 95 === In the standards of the hybrid video coding systems, such as H.264/AVC, H.263, MPEG-2 and MPEG-4, their motion estimations are based on motion compensated prediction with the fractional pel displacement vector resolution(1/2-pel MV resolution). Recent approaches, such as MPEG-4 ACE and H.26L, use the 1/4-pel and 1/8-pel displacement vector resolutions, respectively. In order to achieve estimation and compensation for high- resolution displacement vectors, image signals have to be determined by interpolation process. Thus, interpolation filters are applied in motion compensated prediction systems. In this thesis, we propose a new method to design the interpolation filter. The method enhances the motion-estimation accuracy and the quality of the motion compensated prediction. The proposed method considers the spatial and temporal correlation, and use the Least-Square estimation technique to find the best interpolation filter which filter-coefficients are different for each sequence. After sampling pixels from each frame of a sequence, we can obtain the coefficients of the interpolation filter by Least-Square procedure. The experimental results show that the proposed method provides better quality. Moreover, the Least-Square estimation doesn’t need expensive computation, which prevents decreasing system efficiency.