Reconhecimento de formas em imagens através da associação de pontos e avaliação multicritério de arestas

Abstract

The task of recognizing shapes in images is an activity located in the area of computer vision and is directly related to the problem of pattern recognition. Recognize a pattern in an image can be made possible through the use of graphs for the representation of their features, since the process of recognizing two images can be run through the matching of the graphs that represent them. In this work, images shapes will be represented by graphs and their matching process is modeled as a quadratic assignment problem. In addition, evaluation techniques are proposed to cost the contribution of the vertices and edges associations. We propose two adaptations on the Shape Context technique and also a matching method based on depth search using a multicriteria evaluation on the characteristics of edges. The aplication of these techniques to line graphs generated from the original matching graphs is proposed, with promising results. The techniques implemented here are submitted to tests applied on four image databases used in several papers in the literature.