Research Of Interactive Image Segmentation Algorithm Based On Random Walk

Image segmentation is a very important part of image processing. In order to meet the requirement of complex image segmentation, people propose interactive image segmentation that artificially specifies some information as input in the process of segmentation. In recent years, interactive image segmentation has become a hotspot in research.Based on analyzing the current interactive image segmentation algorithm, Random walk algorithm is emphasized in this paper. Through the simulation experiment analysis, this thesis points out the problems of traditional random walk:running time problem, noise disturbance problem, interactive operation problem and depth concave problem.This thesis proposes an improved random walk algorithm for the problem mentioned. First, it adopts Mean Shift to preprocess the image, partitioning it into a series of homogeneous areas, so that noise is inhibited while reducing the number of nodes. Second, it uses Polynomial Mahalanobis Distance to define the weight between regions, which takes in account not only color and spatial characteristics, but also input data. Thirdly, it adds the auxiliary seeds, and uses the auxiliary and signed seeds to walk random, realizing region merging. Finally, it uses the signed seeds to further walk and adjust, realizing final image segmentation.In order to prove the validity and practicability of the improved algorithm, the experiment is done with natural and medical images, and the results show that the improved algorithm simplifies the interactive operation, enhances the anti-noise performance, and improves the segmentation effect of depth concave regions.