A region-based object recognition algorithm

Abstract

This paper presents a new algorithm for recognition of objects in a given image. This algorithm transforms the input image into a directed graph which is constructed through several defined rules. The graph characteristics represent the global shape information of the object inside the input image, and are extracted during graph construction. This strategy prevents the postprocessing traversing to the graph, with consequent improvement of computational time. The algorithm was tested over a specific database, and the experiments were conducted to show its performance in the light of two types of problem: object class recognition and similar image retrieval.