Saliency Detection Via Object Proposal Selection

With the development of technology,human get a mass of image information fromvarious aspects at each time.How to deal with these digital image quickly and efficiently is becoming a challenge that human must confront.Saliency detection can utilize computer automatic segment some object regions which attract human attention from images.As the pre-processing,saliency detection play an important role in image processing aspects,such as image retrieval,object recognition and so on.Object proposals have very diverse shapes and sizes.Among them,only a small proportion carries the visual cues of objects and helps salient detection.However,existing proposal algorithm produces thousands of proposals and the scoring mechanisms do not work well and usually unable to accurately evaluate the proposals.In this paper,we examine the characteristics of proposals and propose a novel coarse-to-fine framework to detect salient objects.In coarse stage,we re-define the evaluation indicators to describe the saliency of each proposal.We output the top-scoring one and treat it as pseudo ground truth.In fine stage,we train a structural SVM ranker across a group of images to rank the proposals and obtain the proposal-level saliency map for each image through a weighted fusion of top-ranked proposals.Different from traditional rankers,which balance the accuracy of the full list,this ranker prefers the high-quality proposals to be ranked at the top regardless of the rest.After that,we further refine the saliency result by combining the finer processing based on superpixels and get the superpixel-level saliency map.Finally,the saliency result is combined by proposal-level saliency map and superpixel-level saliency map.Experimental results on five benchmark datasets show that the proposed method performs favorably against nineteen state-of-the-art saliency methods,especially the precision-recall curves.In the experiment,we evaluate saliency map from qualitative evaluation and quantitative evaluation.The final result can highlight the salient object accurately.