Contribution of Active Contour Approach to Image Understanding

Abstract

The paper proposes a formal, though a simple, model of the activity of human vision system. That model serves as a reference point for description of relational and contextual recognition and allows to formulate a definition of image understanding problem. As a potential solution of that problem active contour methods are proposed, which do not possess some limitations of classic recognition algorithms. Moreover, they can be used not only as methods of construction of contextual pixel classifiers, but also as methods of identification of relations between any arbitrary more complicated structures (concepts).