Moment-preserving pattern matching

Abstract

An accurate template matching algorithm of low computational complexity that facilitates low-cost hardware design and VLSI implementation is highly desired by many image-processing applications. This paper describes a template matching method, called moment-preserving pattern matching (MPPM), which reduces the computational complexity of template matching significantly and provides several advantages in hardware implementation. The methodology is based on the concepts of pairing function and moment-preserving quantization. The experimental results show that it is computationally fast and accurate in locating the template if noise corruption is not too severe. For obtaining a template matching algorithm of high accuracy and robustness against noise, and low computational cost, MPPM is applied to a two-stage matching scheme. A statistical model for estimating the correlation threshold is also given.