Low-cost and high-performance motion estimation algorithms and architectures

• Main Authors: Huan-Shuan Fu, 傅煥軒
• Other Authors: Shyue-Kung Lu
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/49806345504926325468

碩士 === 輔仁大學 === 電子工程學系 === 98 === In digital video systems, video encoding possesses heavy computation complexities for all the systems. Therefore, video compression technique employs temporal and spatial correlations to achieve reduction of computations and the data amount. Among the video compression techniques, motion estimation plays an important role to remove most of temporal redundancy. The full search block matching algorithm (FSBMA) is the most popular scheme for motion estimation due to its good quality. However, it demands the most computations. This thesis describes three new techniques that can be used to reduce the huge computations of motion estimators. The first technique uses a novel downsampling method to decrease the hardware area for about 33.41%. The second technique improves the multilevel successive elimination algorithm (MSEA) by using the new three steps search (NTSS) as the first step to find a good initial search point. The computation complexity can be reduced for about 31% ~ 34%. The finally technique combines the NTSS algorithm and the global elimination algorithm (GEA) to achieve 83% ~ 85% reduction of computations. However, the video quality will loss significantly.