Unequal-arm adaptive rood pattern search (UARPS) algorithm with accurate initial points

Adaptive rood pattern search (ARPS) algorithm is a famous block matching motion estimation algorithm with a good performance. ARPS compared to diamond search (DS) algorithm is faster with almost similar quality (PSNR). ARPS in addition to four blocks that form arms of the rood pattern, searches the block that is predicted by the motion vector of left neighboring block as initial points. In order to find the best matched block at the first step that will accelerate finding motion vectors and also to use half way stop technique efficiently, more initial points are added according to high coherency between motion vectors of neighboring blocks that will lead to discovering the best matched block faster and accurately. Furthermore, region prediction is applied to the first step of the proposed unequal ARPS search pattern to eliminate the useless points that aren´t in the predicted region and also to prevent algorithm from trapping into local minima. Experimental results show that the proposed algorithm reduces average number of search points significantly compared to the previous algorithms, while PSNR values are closer to full search (FS) algorithm.