Research On Interactive Image Segmentation Method Based On Superpixel

Image segmentation is the most important underlying technology in computer vision.With the development of computer vision technology,it is widely used in image retrieval,medical image segmentation and other research fields.At the same time,image segmentation is also the basis of image analysis,the results will directly affect the quality of subsequent image processing.Therefore,image segmentation has important research value and research significance.As an effective image initial segmentation algorithm,the superpixel preserves the local information of the image at the same time as the fast segmentation of the image,and greatly reduces the complexity of the subsequent image processing.Compared with automatic image segmentation,interactive image segmentation makes it easier to segment image by adding a priori information of user interaction.Therefore,this paper will use the method of interactive image segmentation based on superpixel to segment the image.The main work of this paper is as follows:(1)Analyze and summarize the superpixel correlation theory and technology,and divide the superpixel algorithm into two categories,and introduce the superpixel algorithm which is more commonly used.Among them,Mean-Shift and GrabCut are introduced.By comparing and analyzing the limitations of the existing superpixel algorithm,it is determined that this paper will use the SLIC superpixel algorithm that is used for initial segmentation.(2)Aiming at the segmentation speed of some superpixel segmentation algorithm that is slow,and the extraction edge is not ideal and so on.through the experiment to determine the experimental parameters for SLIC superpixel algorithm set for the initial segmentation of image.And the experimental parameters of the SLIC superpixel algorithm are experimentally determined to be used for initial segmentation.Although the color histogram can effectively represent the color characteristics of color images,but a single feature can not effectively represent the image of other important features of the information,on the basis of a color histogram and Contourlet transform composed of regional similarity characteristics to represent the characteristics of the superpixel region after segmentation.(3)In this paper,based on the MSRM method,a hierarchical matching method based on region fusion is proposed to overcome the false matching of the superpixels in the region fusion.By changing the region's fusion strategy,the maximum similarity principle is used to fuse the superpixel region.The experimental results show that the proposed method not only improves the segmentation speed,but also improves the accuracy of the segmentation(ACC),while reducing the negative rate metric(NRM).