Spatial correlation based fast field motion vector estimation algorithm for interlaced video encoding

Abstract

A fast motion vector estimation algorithm for field motion estimation having computational complexity independent of search range is presented for interlaced video encoding. The algorithm is based on spatial correlation of motion vectors and yields good tradeoff between motion estimation distortion and number of SAD computations. The speed up achieved by the algorithm is in the order of 190-280 times and the SAD increase is 2-10% compared to full search. The number of bits required to code motion vectors is less as the algorithm is based on correlation of motion vectors. Overall performance of the algorithm is compared with full search and three-step search using speed-up, average SAD per macroblock (MB), motion vector bits per MB and PSNR as objective measure. The proposed algorithm can be used in standards such as MPEG-2, MPEG-4 ASP, etc. The proposed algorithm is implemented in our MPEG-2 encoder and performance of the algorithm is presented for the same.