Image Matching based on Relaxation and Model Switching on Contour Characterization

Relaxation labeling algorithms introduced by Rosenfeld et al., have been widely used in image and sequence matching applications. The underlying mechanism of labeling via relaxation is a deterministic dynamical system whose initial status converges to one of the attractors. This is also a process of the input image segments being assigned the predetermined labels in a parallel and unsupervised way. Modeling of images always precede the relaxation stage, however, apparently similar images can be judged as being quite distant, according to the nature of the modeling process. In this work, an approach for contour-based shape matching named as Constructive Relaxation Matching (CRM) is introduced. In CRM, the modeling stage for a novel input image contour, commonly done in the same procedure used for modeling the templates, will be included in the procedure of iterative relaxation matching. Upon dynamically constructing the model during relaxation, pairs of contour objects having similar template label assignment probabilities will be unified to make one object. In the experiments, the method is applied to shape matching problems demonstrating the ability to adaptively model the input image during relaxation, together with an application to similar image retrieval from a pool of images. keywords: Probabilistic Relaxation; Shape Matching; Labeling; Model Switching; Dynamic Systems