Abstract: Owing to traditional methods of image segmentation exist defects in dealing with images of larger gradient changes,this paper integrate inter-pixel gradient differentia information on the basis of original pixel gray information in the random walks to assign weights to edges of figure to better describe relationship of adjacent pixel,by means of the information of seed points designated by user,with the aid of circuit simulation and systemof linear equation to guide the random walkers gradually move from of each of the unseeded points along the edges with maximum probabilities to the right marker point(object or background),which is belonged to the most probable region to carry out the segmentation of image.The experimental results show that this method is fit for processing images with larger gradient changes,which not only retain the anti-noise advantage of original algorithm,but also have higher computing efficiency.To sum up,it's an effective method of image segmentation.