Research On Target Segementation Algorithms Based On Saliency

Image segmentation has a fundamental and critical role as a pre-processing step in the overall image processing.Research on related issues has always received great attention from domestic and foreign scholars.Image segmentation is still a challenging task due to the variety and complexity of the information contained in image.Among them,the saliency-based target segmentation algorithm is an important branch of the image segmentation algorithm.It simulates the human visual perception mechanism and can quickly and accurately segment the interested target or region of an image.Important information can be selected or focused from a large amount of image,which is significant in the field of image segmentation.This paper studies the algorithm based on saliency target segmentation.Firstly,the basic theory and several classical methods of the saliency target segmentation algorithm are summarized.Then the existing methods are analyzed and the existing problems are discussed.Focusing on some existing problems,this paper proposes a new automatic saliency target segmentation algorithm based on support vector machineand a non-convex low-rank saliency target segmentation algorithm based on structural matrix decomposition.The main work and research conclusions are as follows:(1)Aiming at the problem that the inhomogeneous saliency map of the saliency detection algorithm reduced the image segmentation accuracy,a new automatic saliency target segmentation algorithm based on support vector machine(SVM)is proposed.The algorithm firstly obtains the approximate salient region and the background region by saliency detection,quantizes the image in the HSV color space,and uses the color histogram information to obtain the dominant color of the salient region and the background region,thereby automatically selects the positive and negative training samples.Then,the SVM is trained by features,such as color,extracted from the training samples.Finally,the SVM is used to accurately segment the entire image to obtain a significant target.Theoretical analysis and experimental results show that the proposed algorithm not only has the mechanism of correcting the error detection information,but also effectively solves the problem that the segmentation result lacks accurate boundary due to the inhomogeneous saliency map.So,it can achieve more accurate segmented result than the existing similar algorithms.(2)Aiming at the problem that the saliency map of the saliency detection algorithm is inhomogeneous and the target background is confused,a non-convex low-rank saliency target segmentation algorithm based on structural matrix decomposition is proposed.The algorithm is mainly divided into two major steps.The first step is to construct a matrix decomposition model by using a non-convex S_p norm constrained low rank background and an F norm improved sparse constraints,and then extract multiple features of the image to form a feature matrix F,and then obtain the sparse matrix S and the low rank matrix L by the matrix decomposition model.Finally,the saliency map is calculated by the sparse matrix.The second step is to further segment the saliency map using the OTSU threshold segmentation algorithm to obtain a significant target.Theoretical analysis and experimental results show that the proposed algorithm is more complete and uniform than the existing similar algorithms when the background is more complex,and the saliency target obtained by the final segmentation is more accurate.