An accelerated full search based on split-half and backtracking

This paper presents a novel accelerated strategy of full search for traditional correlation matching. Generally, the maximal value in the generated similarity profile is detected, and taken as the correct match position. Obviously, only those peak positions (that is, local peaks) may be available for searching maximal peak when multi-peaks phenomenon exists; however, it will take a lot of computation time for those non-peak positions. Based on the continuous of similarity rates, a split-half method is proposed to reduce the number of checked positions. Meanwhile, for those local peaks skipped by the split-half method, a backtracking strategy is applied to recover it. In this study, the experimental results with synthetics and real world images have demonstrated that the proposed algorithm can reduce about 30% costs without losing estimation precision almost.