Cosalientcy Detection Based On Similar Matrix And Clustering Consistency

Human vision ability can easily identify salient objects and image segmentation,but for computer,it is still a formidable challenge.In recent years,due to the large number of image information and the similar salient objects in multiple images,it is more efficient and reasonable to detect multiple salient images.This kind of salient detection is called co-salient detection of multiple images.Co-salient detect the common salient objects within a group of images,and the salient objects in the individual images,which are considered as saliency non-coordinated parts,will be suppressed as the background.The joint segmentation of multiple images containing similar objects is called co-segmentation.The applications of Co-salient detection are widely used in many fields,such as co-segmentation,co-positioning,image search and so on.The main work of this paper is focused on the following points:1.A co-salient detection method based on similar matrix is proposed.First of all,the images are divided into super-pixels and generated saliency maps with existing saliency detection algorithms,and we perform saliency thresholding to extract the saliency regions for each images,by selecting salient superpixels based on different maps;Secondly,The histogram vector is constructed by the RGB color feature of the saliency region,and all the row vectors are combined into a feature matrix.Then all the histograms should have similarity characteristics and construct the similarity matrix;Thirdly,we use the low rank matrix recovery model to decompose the feature matrix to obtain the noise sparse matrix,and get the weighted value,and by fusing weighted value and the initial salient maps to get salient maps.2.In order to solve the problem of non-cooperative salient objects suppression in traditional algorithms,a new co-salient object detection method based on clustering consistency is proposed.The clustering consistency is used to obtain the co-salient value,and the weighted salient values are fused,then the final salient map is obtained.3.Based on the co-salient detection method,a co-segmentation model is proposed to realize the co-segmentation of multiple images.First,extract the curve contour of the salient region of co-saliency detection method;Then,the contour contour is evolved by active contour,level-set and energy function optimization.Finally,the co-segmentation results of the final image are obtained.4.ICosegSub and iCoseg databases are used to verify two co-saliency detection methods in this paper.Experiments show that the two methods in this paper achieve high accuracy.Compared with the traditional algorithm,the algorithm significantly suppresses non-cooperative salient regions,and improves the accuracy of co-saliency detection.ICoseg and MSRC databases are used to verify the co-segmentation method,and the experimental results show that the proposed co-segmentation has high segmentation accuracy.