A Novel Image Segmentation Method Based on Random Walk Combined with Robust Sampling Prior

In the image segmentation technique, how to accurately solve the boundary of foreground objects and background is the key problem, and the typical unsupervised interactive binary segmentation algorithms in this include Grab Cut, random walker and K-means clustering, etc. Among them, the random walker algorithm is a classical algorithm of segmentation based on graph theory, which has the advantages of strong interpretation and high efficiency and is suitable for segmentation fields with low boundary requirements. However, because the random walk algorithm only uses the neighborhood information of unknown pixels and does not make full use of the sample prior information at longer distances, the results of this method are segmented more coarsely. This paper proposes a random walker image segmentation method based on robust sampling. The method selects the sample points that are far from the unknown pixels for robust sampling and selects the best foreground and background sample pairs for the calculation of the foreground prior probability. After that, the computed prior probability values are fused with the random walker algorithm as a way to improve the accuracy of the segmentation results. The experimental results show that the algorithm of the addition of prior information can obtain better segmentation results than the original random walker segmentation. The experiments in this paper compare the mean square error values of some typical images to further validate the effectiveness of this method.