Automated region segmentation using attraction-based grouping in spatial-color-texture space

A great deal of attention has been paid to image content-based retrieval systems (CBRS). One important goal in CBRS is to extract local low level image features such as color, texture and shape, to allow queries based on these features. A large CBRS containing tens of thousands of images requires an automatic feature-extraction method since human aided segmentation is impractical. We address this problem in a particular application setting by using an attraction-based grouping method in spatial-color-texture space. The attraction concept makes this approach similar to human aided segmentation. Experimental results show that the method is reasonably better than existing methods, and has the potential to be used in other CBRS-related applications.