An object class-uncertainty induced adaptive force and its application to a new hybrid snake

Object segmentation is of paramount interest in many imaging applications, especially, those involving numeric, symbolic, syntactic, or even high level cognitive knowledge perception. Among others, “snake”—an “active contour” model—is a popular boundary-based segmentation approach where a smooth curve is continuously deformed to lock onto an object boundary. The dynamics of a snake is governed by different internal and external forces. A major limitation of the present framework has been the difficulty of incorporating object-intensity driven features into snake dynamics so as to prevent uncontrolled expansion/contraction once the snake leaks through a weak boundary region. In this paper, a local-intensity-driven “adaptive force” is introduced into the model using object class-uncertainty theory. Given a priori knowledge of object/background intensity distributions, class-uncertainty theory yields object/background classification of every location and establishes its confidence level. It has been demonstrated earlier that confidence level is high inside homogeneous regions and low near boundaries. In the current paper, object class-uncertainty theory has been applied to control snake deformation leading to a new adaptive force acting outward (expanding) inside intensity-defined object regions and inward (squeezing) inside background regions. It has been demonstrated that the method possesses potential to resist uncontrolled expansion of a snake contour (for an expanding type) inside background after leaking through a weak boundary. Further, it has been shown that the adaptive force operates in a complementary fashion with the image intensity gradient by reducing its strength near boundaries using the confidence level of classification. Another major contribution of this paper is the formulation of a “hybrid snake” (HS)—a new model, where an initial contour is gradually deformed over a hybrid energy surface composed of some direct energies (e.g., internal energies) and other indirect energies contributed by local contour displacements over a force-field (e.g., image or user-constrained force-field). Applications of the proposed adaptive force-enabled HS on different phantom and real images have been presented and comparisons have been made with a conventional snake (CS). Finally, a quantitative comparison based on computer-generated phantoms at various levels of blur and noise has been provided.