A Novel Framework for Object Recognition under Severe Occlusion

Abstract

This work introduces a novel technique for the automatic identification of real world objects in complex scenes. The identification problem requires the comparison of assemblies of image regions with a previously stored view of a known prototype. Shape context representation and matching are employed for recovering point correspondences between the image and the prototype. Assuming that the prototype view is sufficiently similar in configuration with an object in the complex scene, the correspondence process will succeed. However, a number of ambient conditions such as partial object occlusion and contour distortion, may affect the performance of the matching process and consequently the identification result. A novel multistage type of clustering of suspicious image locations is applied in a novel fashion to enable the identification of regions of interest on the complex scene, based on a set of density and figural continuity metrics. In order to increase the robustness of the identifier to mismatches and reduce the computational cost of the process, a selection of the initial suspicious regions is applied. The performance of the identifier has been examined in a great range of complex image and prototype object selections.