Research On Target Detection Based On Co-saliency

With the rapid development of Internet technology and digital multimedia technology,massive amounts of image data are generated every day.Since the human visual system has the ability to quickly locate salient target in an image,the region of interest in the image can be quickly processed.Salient target detection is to mimic the working mechanism of the human visual system and quickly locate the salient regions in the image.It can be used as a pre-processing step for subsequent advanced tasks.Therefore,the study of salient target detection has received more and more research scholars' attention.Aiming at the limitations and shortcomings of the existing salient target detection methods,we can get the strong estimation of the image saliency target by combining different weak saliency detection methods based on different detection principles.The detection result is more accurate and more applicable.The main contents and contributions of this paper are as follows:1?We propose a single image salient target detection algorithm based on co-saliency.Firstly,we block the detected image and a plurality of weak saliency detection algorithms are used to detect the salient target.Then,we can obtaine a plurality of saliency detection maps,and get the feature matrix by the plurality of weak saliency maps and the super pixel segmentation result.Next,we get the saliency value of each super pixel block by the calculation of the energy function with the feature matrix.We select the seed points by comparing the super pixel block saliency value with the threshold value and establish a weight matrix according to the relationship between the pixel points of the detected image.Finally,we constructe the migration equation by the weight matrix and the saliency values of other non-seed points are obtained by the seed point migration to get the final saliency map.Experiments show that our method can detect the salient target in the image more accurately and performance well on different data sets.2?We propose a multi-image co-salient target detection algorithm based on co-saliency.Firstly,we make full ues of the inter-correlation information between multiple images and cluster the pixels in multiple images by the features.Then we calculate three salient clues based on the clustering results which are contrast clues?spatial clues and correspondingclues.After obtaining the initial saliency map of all images,we can get the co-salient target detection result by optimizing the initial saliency map with the weight by the three cues.Experiments show that our method can detect the co-salient target between multiple images more accurately.