Self-organizing neural networks based on spatial isomorphism for active contour modeling

Abstract

The problem considered in this paper is how to localize and extract object boundaries (salient contours) in an image. To this end, we present a new active contour model, which is a neural network, based on self- organization. The novelty of the model consists in exploiting the principles of spatial isomorphism and self-organization in order to create flexible contours that characterize shapes in images. The flexibility of the model is effectuated by a locally co-operative and globally competitive self-organizing scheme, which enables the model to cling to the nearest salient contour in the test image. To start with this deformation process, the model requires a rough boundary as the initial contour. As reported here, the implemented model is semi-automatic, in the sense that a user-interface is needed for initializing the process. The model's utility and versatility are illustrated by applying it to the problems of boundary extraction, stereo vision, bio-medical image analysis and digital image libraries. Interestingly, the theoretical basis for the proposed model can be traced to the extensive literature on Gestalt perception in which the principle of psycho-physical isomorphism plays a role.