Abstract: Image segmentation and contour detection is important for computer vision and pattern recognition, and the active contour for image segmentation is treated as the most popular study hotspots. A detailed analysis of Mumford-Shah model is given with emphasis on its active contour mechanism, and then a piece-wise smooth level set model is discussed. By constructing the new soft distance function to reinitialize the level set function during the iterations, and by combining the essentially non-oscillatory scheme and the predict-revised scheme, a new finite difference algorithm is proposed. This algorithm not only can detect multiple objects with various shapes of topology including convex and concave objects, but also has the ability to preserve the gray property of the multiple objects. Finally, some examples are given and it is shown that the new algorithm is stable and oscillation does not appear.