Research On Aotomatic Image Segmentation Algorithm Based On Grab Cut

With the rapid development of electronic technology and digital image technology in recent years,as the key and foundation in the fields of image comprehension,pattern recognition and computer vision,image segmentation technology has very important research significance.All kinds of image segmentation algorithms are very diverse and intricate.Among them,Grab cut algorithm,as one of the efficient,accurate and widely used interactive color image segmentation algorithms,has been a hotspot studied by scholars in recent years.Firstly,this paper has analyzed the principle of Grab Cut algorithm and several aspects that need to be improved.The algorithm is developed from Graph cuts algorithm,combine graph theory technology with optimization strategy.Its basic idea is to map the image into a corresponding network weighted graph,and by this way,the image segmentation problem to be solved is transformed into the problem of the minimum cut of the graph.Aiming at the shortcoming that Grab Cut algorithm requires user interaction rather than automatic segmentation,this paper propose a method.The salient region image which is obtained using the improved salient region detection method,replaces the rectangular frame draw by users,to initialize the parameters of the two Gaussian mixture model.Since each single salient region detection method has its own advantage and disadvantage,this paper calculates the saliency values respectively from the aspects of global contrast,local contrast,rarity of pixel features,and then linearly fuses the three salient region image into one salient region image according to their each proportion of salient points,so as to better play their respective strengths.In the end,a salient region image with better results was obtained.Then the appropriate threshold is selected to binarization the salient region image,which is used for pre-segmentation,so that the original algorithm becomes automatic segmentation algorithm.Aiming at the problem that Grab Cut algorithm has a high error rate when the foreground and background of the image have similar colour,this paper proposes an improved method.The preceding improved salient region image is added to the data penalty item of the energy function with certain weight as a constraint item,so as to improve the reliability of data item constraint in complex background.Because Grab Cut algorithm mainly relies on the region information and lacks the constraint of edge information,the target edges extracted in some cases are not smooth and accurate enough.Aiming at this shortcoming,an improved method is proposed in this paper.The edge information acquired by the edge detection of the image by Sobel edge operator is added to the smooth penalty item of the energy function with certain weight,in order to improve the reliability of the smooth item constraint and improve the quality of the segmentation edge.Experimental results show that this algorithm not only realizes the transformation from interactive segmentation to automatic segmentation,but also improves the accuracy of segmentation results(quantitatively evaluated by recall and precision)and the smoothness of edges.