New Sorting Technique on Partial Distortion Search Using Two Bit-Transform for Fast Optimal Motion Estimation

This paper proposes a novel two bit-transform sorting technique to get a better matching order on the loss less partial distortion search (PDS) algorithm. By sorting the level difference sets calculated from the approximate distortion between the two-bit transformed reference block and candidate block, the matching order can be computed to apply to the typical PDS. To reduce the checked pixels further, the initial minimum sum of absolute difference (SAD) is derived from the cardinality of the level difference sets among candidate blocks. The experimental results indicate the proposed algorithm reduces computational complexity by about 27% on average compared with the typical PDS in terms of the average number of the checked pixels for night benchmark sequences.