Research On Normalized Cuts Based On Saliency Detection

Image segmentation is an important computer technology in the field of digital image processing and computer vision.It plays a very important role in the current field of computer vision related fields such as target recognition,medical image segmentation,and driverless driving.The normalized cuts algorithm based on graph theory is the most popular image segmentation method developed in the past ten years.It has achieved good results in various experiments and a very ilImportant influence in various applications.As an algorithm for solving the global optimal solution,it has the advantage of not requiring preprocessing and it is not easy to produce discrete points.However,the normalized cuts algorithm has low segmentation accuracy,is easily subject to background features,and has a large spatial complexity,is an NP hard problem.Therefore,based on the analysis of the advantages and disadvantages of the traditional normalized segmentation,this paper preprocesses the saliency detection of the weight matrix prior to the calculation process of the weight matrix,and obtains several salient region blocks in the image,and then according to the significant The approximate relationship betxween sex regions calculates the weight for each point of the image.Then we solve the generalized eigenvalue problem according to the relevant conclusions of Rayleigh quotient,and use the feature vector corresponding to the second smallest eigenvalue to segment the graph,so as to realize the image segmentation.A large number of experimental results show that the segmentation algorithm in this paper can effectively improve the image segmentation results,and can achieve better image segmentation results in more complex backgrounds.The main work of this article is as follows:(1)First of all,this paper proposes a normalized segmentation algorithm based on saliency detection.For normalized segmentation,it is easily affected by backaground features.This aigorithm effectively improves the accuracy of target segmentation and positioning.The fusion method of the saliency detection and the nornmalized segmentation algorithm is given:in the construction of the weight matrix,the importance of the salience points is assigned to the weight map,and the weights are reconstructed.(2)Secondly,we improve the method of saliency detection,and proposed a saliency detection method based on multi-scale.We use multiple scales to reduce the effect of recurring global background features,allowing the algorithm to segment more complex background images.By experimental results,it is confirmed that the proposed method can significantly improve the accuracy and application scenario of the normalized segmentation algorithm.