Research On Image Segmentation Algorithm Based On Superpixel And Graph Theory

Image segmentation divides the original image into several disjoint regions,and extracts the required target objects from the image.Image segmentation is not only a basic step in the image processing process,but also a more critical step.The quality of image segmentation will affect subsequent image processing effects.Because there is a one-to-one correspondence between the vertices in the graph structure and the pixels in the image,the image segmentation method based on graph theory has received widespread attention.When the image has fewer pixels,the segmentation method based on graph theory can get better segmentation.As the number of pixels gradually increases,the relationship of the graph structure becomes more complicated,which greatly increases the computational complexity.Therefore,before segmentation using the graph theory method,the image needs to be pre-processed to generate several pixel blocks.This paper studies image segmentation methods based on superpixel and graph theory,and proposes some improved algorithms,which mainly include the following aspects:Aiming at the shortcomings of GrabCut's iterative process,long running time,and mainly relying on region information,lack of edge information,and inaccurate segmentation results,this paper proposes an improved GrabCut algorithm combined with superpixels.First,simplify the operation steps of the SLIC algorithm and directly specify the approximate number of pixels contained in each superpixel,so that the same parameter can be applied to multiple images with different resolutions.Then use the SLIC algorithm to pre-segment the image,generate superpixel blocks,and use the superpixel blocks as vertices to build a simple network graph.Finally,the Sobel operator is introduced into the smoothing term,a new smoothing term and energy function are defined,and subsequent segmentation of the image is performed using the GrabCut algorithm to output the final segmentation result.The experimental results show that the improved GrabCut algorithm reduces the segmentation time to a certain extent,and further improves the image segmentation effect while ensuring the original algorithm effect.Aiming at the problems of the GrabCut algorithm in segmenting pictures with complex backgrounds and similar background and foreground colors,the segmentation effect is poor and the segmentation accuracy is not high.A graph-based and GrabCut image segmentation algorithm is proposed.First,set the k value in the Graph-based algorithm to the sum of image length and width to improve the segmentation efficiency of the algorithm.Then,the image is pre-processed,and the image is initially divided into different blocks using a Graph-based algorithm.Next,carry out manual interaction.The user marks the rectangular frame,calculates the pixel labels in the rectangular frame,and calculates the proportion of each label in the entire picture.Mark some pixels as the background by some judgment criteria,reduce the interference of some background pixels,and update the foreground and background pixel labels.Finally,the GrabCut algorithm is used for subsequent segmentation of the image to obtain the final segmentation result.The experimental results show that the improved algorithm has certain advantages in processing pictures with more complex backgrounds and similar background and foreground colors,and has better segmentation results.