A Study On Co-saliency Detection

With the rapid development of the computer networks and the technology ofmultimedia, multimedia data like digital pictures is blowing up quickly. It’s a greatproblem to deal with these problems and use them. A series of technologies forprocessing images in the field of computer vision,(image retrieval, imagesegmentation, object detection and co-segmentation of images, etc) were raised.Although these methods have achieved many good results in the treatment of thedigital image, they are too complex in the aspect of time, due to the huge quantity ofmultimedia data. In the same time of the optimization of algorithm to reduce theprocessing time, it can also reduce greatly the processing time if it only processes.Therefore, it’s a urgent problem to detect rapidly and accurately the area withprominent object of a picture.The task of co-saliency detection aims at automatically discovering the commonsalient regions existing in multiple images. Although several saliency detectionmethods have been proposed in recent years, they have respective pros and cons.This paper presents a general fusion scheme to integrate multiple saliency/co-saliencymaps, which results in an adaptively weighted co-saliency detection better than eachconsidered individual saliency map. Specifically, the fusion is based on a simplelinear combination of multiple co-saliency maps with adaptive weights. We propose ageneral consistency criterion to automatically determine the optimal weight of eachindividual saliency map. The proposed method is general enough to combine almostall kinds of saliency detection methods. Experimental results on benchmark datasetshave demonstrated the satisfactory performance of the proposed method onco-saliency detection.