Minimizing the energy of active contour by using a Hopfield network

A Hopfield network for minimizing the constrained energy of an active contour model (snake) is proposed. Taking advantage of the parallel computation and energy convergence capabilities of the Hopfield network, this method is faster and more stable for resolving the boundary locating problem than the conventional methods. The Hopfield network is superior to conventional methods in three ways. First, the Hopfield network can be implemented in a parallel architecture, instead of the sequential process, in real-time application. Second, it guarantees that the energy of the snake can converge to a minimum stable state. Finally, it allows hard constraints to be included as a part of the minimization process. As a result, the present study reveals the possibility that a human vision system can locate an object boundary by using a mechanism similar to the active contour model