Image segmentation based on level set method

In this paper, A segmentation model that combines techniques of curve evolution, the Mumford-Shah model and level set method was presented, to detect the contour of object in a given image, the model can detect object whose boundary is not necessarily defined by gradient and whose gray structure may be complicated. First we construct signed distance function, adopted a method which based on the times that is odd or even numbers through close curve from the point along a direction (if need, may be along several directions) to construct sign table. Then we used improved Mumford-Shah model to segment image, we consider that the object to be segmented is made up of some different gray level, it is difficult to detect the object contour using the Mumford-Shah model, for general objects, the contour of the object is piecewise-contour of along the edge, and the gray difference among the object points nearby the contour is little, so we divide the curve into finite segment, compute gray average of narrow band in and out of the curve, and compute the gray difference between the inner narrow band and outer narrow band of the curve, using improved Mumford-Shah model to segment the object. Experiment results show that the proposed algorithm can be used to segment object without edge and with complex gray structure, and the performance of the algorithm is satisfactory.