Contour Detection Based on Self-Organizing Feature Clustering

The real vision system has a well-developed ability to detect multiple contours and recognize various objects in images. Previous simulation models to perform this process often employ image segmentation or contour integration algorithms. In this paper a new model is proposed to separate individual object contours from the background by the feature clustering. The model is inspired by the contrast mechanism and the self-organizing characteristic of the vision system. It can group edge elements with similar local features together automatically. The self-organizing map (SOM) is used in the model to classify the edge elements in the image. Experimental results show that the object contours can be separated effectively by this model. The model can be used to supply useful information to higher-level visual mechanism for better object recognition.