A Novel Fast Template Matching Algorithm Based on Bounded Approach Using Block Partitioning Method

Exhaustive search template matching in real time object detection requires a large computational time. Reducing the number of computation for full search template matching is the current research interest. This paper presents a fast and efficient template matching method that uses difference of sum norms (L2 norm) as bounding measure that rapidly eliminates large number of mismatching candidates identifying only small number of locations on which the computation of normalized cross correlation (NCC) function is carried out. This method ensures high degree of detection accuracy and furnishes the same result as the full search NCC algorithm. Experimental comparison has been done with existing fast methods using number of real image data sets consisting of small to very large size images with number of template images varying in size. The results show that the proposed algorithm is significantly faster than the other fast matching algorithms.