Fast algorithms for the estimation of block motion vectors

Abstract

The most important fast block matching algorithms are analysed and evaluated. Then a new fast search method, the "Spiral Search Algorithm" (SSA), is introduced. It is a three step algorithm which follows a spiral path searching outwards for candidate locations that satisfy the matching criterion. The efficiency of the SSA arises from: (1) the reduction of the candidate locations without leaving out zones of pixels where the mean absolute difference is not evaluated, and (2) the reduction of computations since many candidate locations are being bailed out. A comparison of fast search methods and the Full Search (FS) approach is presented for a number of video sequences. The SSA is proven to be an excellent compromise between quality and speed. The hardware implementation of the block motion estimation algorithms is also discussed.