Automatic generation of initial contours on image segmentation

In this paper, a new initial generating method is proposed for the level set-based segmentation approaches. An energy model is introduced based on the piecewise smooth Mumford and Shah (MS) model. To solve the two unknown diffusion functions in the model, some linear and nonlinear filtering techniques are applied. They are approximately represented as ldquoforegroundrdquo and ldquobackgroundrdquo components of the image, which can be approximated by diffusion versions of the image, respectively. An initial segmentation can be obtained as the minimum of the energy. Thus, since the initial contour has been denoted as level set function, it can be used as an initial curve integrated with other level set-based segmentation approaches. As a result, the curve evolving paths are greatly shorted, the convergence is speeded up. Some numerical results are presented to demonstrate the effectiveness of the proposed method.