An edge-based optimization method for shape recognition using atomic potential function

Abstract

The edge potential function (EPF) approach is a promising edge-based shape matching tool for visual target recognition, and describes the similarity between contours by means of a potential field. However, background noise in test images may degrade the accuracy of the EPF approach in the identification of target contours. Furthermore, the computational load of the EPF approach is usually heavy, thus limiting its use in online applications. To solve these problems, this paper proposes a new shape matching tool based on atomic potential function (APF). The APF approach reduces the effects of background noise by introducing the concept of atom potential to the generation of potential fields. Moreover, in our proposed APF approach, the potential field is calculated using the contour extracted from a pre-defined target template rather than contours extracted from test images. Following the calculation of the potential field, the derived potential field is transformed to match the contours extracted from the test images. The search process for the transformation that matches the contours most closely is modeled as an optimization problem solved by a modified version of the artificial bee colony (ABC) algorithm - the internal feedback ABC (IF-ABC). Compared to the conventional ABC algorithm, IF-ABC effectively avoids premature convergence and significantly improves convergence speed. Experimental results verify the feasibility and efficiency of our proposed APF approach by comparing it with the traditional EPF method. © 2014 Elsevier Ltd.