An adaptive speed function of level set method for moving object extraction

The convergence and stability of the level set method depend on the speed function. Therefore, it is important to define the speed function in a manner that is suitable for the individual application. In the present paper, we propose a novel speed function of the level set method for moving object extraction from a video sequence with a stationary background. The speed function focuses on the precision of moving object extraction. In the proposed extraction method, the outline between moving object regions and the background is estimated in advance based on the Gaussian noise distribution in the frame. The speed function is changed adaptively using the obtained outline in order to improve the convergence and precision of extraction. In addition, a new energy term in the direction of the contour of an object is incorporated into the speed function. The precision of the moving object extraction method using the proposed speed function is evaluated through computer simulations.