Skip to Main Content
Past research work has shown that the process of shape matching can be rendered into an optimisation problem that determines, based on evolutionary algorithms, the best matching score between pairs of object boundaries. This important finding has enabled near planar objects to be identified efficiently when they are captured under different camera viewpoints. Among other evolutionary techniques, the genetic algorithm (GA) has demonstrated its feasibility in matching silhouette images of objects that are captured under a well-controlled environment. As the latter is not guaranteed in practice, the method has also been extended to match fragmented and incomplete contours. Despite the moderate success achieved, the overall performance is rather inconsistent and also varies significantly among different geometries. To overcome this problem, two variants of a novel approach based on the integration of a simple GA, the distance transform and the migrant principle are developed and presented. Experimental results reveal that the proposed methods are capable of matching incomplete and broken contours with a high success rate and exhibit good stability in performance.