Image Co-saliency Detection And Application Based On Energy Optimization

Along with the popularity of devices such as cameras,images appear in groups in the public view,and these images often contain the common salient objects of the same category.The purpose of image co-saliency detection is to detect salient objects of the same category from multiple images.When the foreground region color and the background region color are similar,most existing image co-saliency detection algorithms are susceptible to background region noise and cannot uniformly highlight the co-salienct objects.For the problems in the above image co-saliency detection algorithms,a co-saliency detection based on energy optimization(CSEO)algorithm are proposed in this thesis.The main work of this thesis includes:In order to solve the problem that the co-saliency map is susceptible to background noise,the CSEO algorithm proposes a quality assessment score to sort the saliency map and select simple images in the image group.The quality assessment score includes the separation measure and the consistency measure,and the separation measure is used to evaluate the degree of separation of the foreground region salient value and the background region salient value;the consistency measure reflects the non-cosalienct objects in the image to the cosalienct objects impact.After sorting the quality assessment scores,simple images is selected from the image group to guide the co-saliency detection of other images in the image group.In order to solve the problem that the co-saliency map can not uniformly highlight the co-salienct objects,the CSEO algorithm proposes an energy equation to fuse the single image saliency cues and the inter-saliency cue between the images;Secondly,the CSEO algorithm improves the image boundary connectivity,and introduces the feature difference term on the basis of the boundary connectivity to construct a single image spatial position cues to avoid the large-area missing of the co-salienct objects when the salient objects is large and at the image boundary position.Inter-saliency cue between images reflect the consistent similarity of foreground regions in the image,avoiding the influence of non-cosalienct objects on cosalienct objects.The CSEO algorithm is validated on the iCoseg and MSRC datasets,and the CSEO algorithm is compared with the current popular image co-saliency detection algorithm,and subjective comparison and objective comparison are performed.The results show that the CSEO algorithm can uniformly highlight the co-salienct objects,which can avoid the influence of background noise on the co-salienct objects and the phenomenon that the cosalienct objects is missing when the co-salienct objects is large and at the image boundary position.In order to further verify the performance of the CSEO algorithm,image segmentation is performed using the co-saliency map of the CSEO algorithm and the co-saliency map of the current popular image co-saliency detection algorithm,and the segmentation results are compared on the iCoseg dataset and the MSRC dataset.The results show that the co-saliency map using the CSEO algorithm can segment the co-salienct objects more completely.