Improvement of convergence and stability in moving object extraction by the level set method

Semantic object extraction from a video sequence is an indispensable technique in new content-based applications, such as in the international standards MPEG-4 and MPEG-7. In the present paper, we propose a technique that extracts the shape of moving objects from a video sequence with a stationary background by the level set method. In the proposed method, two concepts are incorporated into a novel speed function of the level set method in order to improve convergence and stability. The first concept is the object map, which represents the outline of object regions and the background. The speed function is changed using the object map in order to improve convergence. The second concept is the contour potential energy, which represents the energy in the direction of an object´s contour. The efficiency of the proposed method for moving object extraction is demonstrated through computer simulations.