On Analytical Study of Self-Affine Maps

Self-affine maps were successfully used for edge detection, image segmentation, and contour extraction. They belong to the general category of patch-based methods. Particularly, each self-affine map is defined by one pair of patches in the image domain. By minimizing the difference between these patches, the optimal translation vector of the self-affine map is obtained. Almost all image processing methods, developed by using self-affine maps, take advantage of either the attracting or repelling behaviors which have been, only, experimentally investigated. In this paper, we analytically study the properties of self-affine maps and prove their attracting and repelling behaviors. Furthermore, the new corner/edge pointing behavior is also proposed for contractive self-affine maps. We show that the conventional cost function of self-affine maps may cause critical uncertainty due to providing multiple equivalent optimal translation vectors. Thus, a new cost function is suggested to effectively tackle this problem. For evaluation, it is used with the selfaffine snake (SAS) for contour extraction. Experimental results demonstrated that the enhanced SAS provides better performance compared to a number of different active contour methods in terms of both solution quality and CPU time.