A novel cellular search algorithm for block-matching motion estimation

Abstract

A novel cellular search (CS) algorithm for block-matching motion estimation is presented. Two different search patterns, namely the large CS pattern (LCSP) and the small CS pattern (SCSP), are employed to perform a search for the best-matching block. The LCSP assumes that the best-matching block can be located in any direction from the centre of the LCSP, and we show that the number of blocks via LCSP searching is less than via other algorithms. Following the LCSP search, the SCSP is used to search those blocks near the centre block. We show that the CS algorithm is computationally efficient; it requires less computation time than other algorithms, such as the three-step search of T. Koga et al. (1981), the new three-step search of R. Li et al. (1994) and the four-step search of L.M. Po et al. (1996).