Fast template matching based on multilevel successive elimination algorithm

Abstract

This paper proposes a fast template matching method based on multilevel successive elimination algorithm (MSEA) and yields exactly the same result as the exhaustive search algorithm. The algorithm uses two stage searching method to effectively determine the peak point of correlation accurately and robustly. In the first stage, the pixel positions are sampled to form a rectangular or hexagonal grid structure of pixels. The SEA or MSEA is performed on the sampled grid positions. The output of the first stage method is the coordinate at which we get the minimum SAD value. In the second stage, centering on this position, we perform the normalized cross correlation (NCC) with box filtering operation on each pixel position in a window size of 16×16 pixels. The best matching point is obtained at the peak point of NCC value. The method has broad applications in the fields of real time moving object tracking, pattern recognition, machine vision etc. Experimental results are presented here to verify substantial computational savings of the proposed algorithm in comparison with the SEA/MSEA.