Research On Interactive Image Segmentation Based On Superpixels And Graph Cuts

Image segmentation is a technique that divides an image into multiple regions that do not overlap each other by marking pixels of different features.Image segmentation based on graph cuts has attracted much attention in the field of image segmentation,because it can efficiently use local information and global information in image,and unified graph cut theory into an image segmentation framework.GrabCut algorithm is the most widely used in all graph cuts algorithms.With the arrival of big data,the image to be processed becomes more and more complicated,so the issue that the efficiency and accuracy of segmentation is reduced was exposed.In order to solve the above issue,this thesis studies the image segmentation framework based on superpixels and GrabCut algorithm,and mainly improves from the following aspects:Focusing on the issue that GrabCut algorithm is difficult to manually initialize and the computational efficiency of it is not ideal when there are complex background or detail rich region in the image,a GrabCut image segmentation algorithm which combines saliency information and superpixels is proposed.First,an improved simple linear iterative clustering(SLIC)algorithm is proposed,the proposed algorithm introduces the gradient influencing factor into its distance formula and improves the step of original SLIC algorithm by incorporating the over small superpixels.Second,extends the improved SLIC algorithm into the manifold ranking algorithm,and initializes the GrabCut algorithm through the saliency map obtained by the manifold ranking algorithm.Then,constructs the GrabCut network flow graph model with the superpixels generated by the improved SLIC algorithm.Finally,uses the max-flow algorithm for segmentation.Experimental results show that the proposed algorithm achieves the function of automatic segmentation and reduces the segmentation time while ensuring the accuracy of the original algorithm.In order to solve the problems of low accuracy of the GrabCut algorithm when there are similar colors between foreground and background or shades in the image,a GrabCut automatic image segmentation algorithm that combines depth information is proposed.The proposed algorithm is based on the improved GrabCut algorithm which combines saliency information and superpixels.First,the GrabCut algorithm is initialized by using the saliency map integrated with the depth information,and then the depth information and saliency information are combined into GrabCut's color model through adaptive weights to improve the energy formula of the GrabCut algorithm.The experimental results show that the proposed algorithm can effectively segment complex images and has higher segmentation accuracy.