Fast full search motion estimation algorithm using MNPDS
Jin‐Soo Kim, Myung Hoon Sunwoo · 2008
This paper proposes the modified normalized partial distortion search (MNPDS) algorithm. The efficient memory reuse (EMR) and the adaptive search range (ASR) algorithms can reduce computation complexity and memory access. MNPDS can reduce computation complexity while maintaining the image quality very close to that of full search (FS). In addition, EMR can reduce the memory access through utilizing the position similarity of the predicted motion vector (MVp) among neighboring Sub-MBs (macro blocks). Moreover, ASR can reduce the memory access through adjusting search range. The proposed motion estimator can reduce computation up to 80% and memory access up to 84%, compared with FS, whereas PSNR decreases only about 0.045 ㏈.