Rotation-Invariant Fast Template Matching Based on Sequential Monte Carlo

Abstract

Template matching is widely applied in Computer Vision. In the case of a template rotation application, it is still nontrivial to find a template matching method with satisfactory matching accuracy and computational complexity. In this work, we propose a fast template matching method based on Sequential Monte Carlo. The method treats the matching process via a Hidden Markov Model(HMM) which establishes a Bayesian framework providing an approximated solution by an importance sampling approach. This solution is utilized to match the template and estimate the position of target template in a background image. Experimental results show a promising template matching improvement in both matching accuracy and matching time.