Research On Interactive Image Segmentation Algorithm Of Graph Cut Model Based On Multi-layer Graph Constraint

With the rapid development of image technology and image application technology,image processing has become more and more demnding,and image segmentation has been extensively studied and popularized in recent years as an important basis for image processing and image analysis.Due to the impact of uneven gray distribution,low contrast and texture features in the target images,traditional image segmentation cannot meet the needs of increasing applications.In this paper,an interactive graph cut algorithm based on multi-layer graph is proposed,which combines multi features and graph cut methods.Based on the interactive information,the user can directly input the interactive information to guide the segmentation quickly and accurately.The model is established by combining several kinds of features and graph cut algorithm.The weights are calculated by the LLGC algorithm.Finally,the segmentation results are obtained by optimizing the energy function according to the maximum flow/minimum cut.Our work mainly includes the following parts:(1)A three-layer super-pixel based graph cut algorithm is proposed.Firstly,by utilizing multiple super-pixel layers in the graph cut algorithm,the super-pixel relations are svpplemented into the graph model,and hence the high-order information is introduced.Not only the color characteristic of the pixel is used,but also the region characteristic of the super pixel is considered in this method,which can provide more information for image segmentation.Through the multi super-pixel layers,over-segmentation can be inhibited,and the super-pixel achieved with different parameters can increase the robustness of the algorithm which enhances the stability of the algorithm.Secondly,the utilization of the LLGC algorithm can make sure that each node passes its local spatial information to its neighbors iteratively until all nodes achieve stability.Because the convergence of the diffusion can be obtained through the derivation,the proposed method is very efficient.Finally,segmentation experiments are performed on Berkeley and Microsoft Research Cambridge(MSRC)image databases.The segmentation algorithm is evaluated by the difference experiment,and the effectiveness and feasibility of the algorithm are proved on the images with complex background and low contrast.(2)An interactive image segmentation algorithm is designed based on image patch layer and super pixel layer.Based on the graph cut method,a method of combining the image block layer and the super pixel layer is proposed.Firstly,image preprocessing is performed to obtain image patch layer based on image patch and super pixel layer based on super pixels.Image patch is used to describe the image local area informtion by dividing the image with a specific window size.Since the mean value of the pixel feature in the image patch is used as the feature information,the texture and structure features are provided.At the same time,the image patches are weakened in the method of keeping the inner edge information of the image.Therefore,the image patches combined with the image layer,the feature information of the region part is enriched,and the disadvantage of the pixel patch to the edge segmentation is also compensated.Then we add the super-pixel layer and the super-pixel relation when we build the graph model to introduce the high-order information.This method can provide more information,and reduce the split error rate.The segmentation experiment is performed on the test images provided by two common image datasets,and the segmentation algorithm is evaluated by the difference experiment,which shows the effectiveness and feasibility of the algorithm in the segmentation of images with complex backgrounds and low contrast.