Saliency Detection Via A Unified Generative And Discriminative Model

Humans have the capability of directing the attention to regions of interest(ROI) rapidly and exploiting the limited resources to process the ROI in complex scenes. Saliency detection, which is a branch of image processing in computer vision, aims to identify the salient object by simulating human visual system and generate a saliency map where the intensity of the pixel indicates the likelihood of it belonging to the salient object.In this paper, we propose a visual saliency detection algorithm which incorporates both generative and discriminative saliency models into a unified framework. First, we utilize the algorithm of searching image patches to collect abundant background superpixels. The background dictionary could be learnt from those background superpixels. Then we develop a generative model by defining image saliency as the sparse coding residual based on a learnt background dictionary. Second, we introduce a discriminative model by solving an optimization problem that exploits the intrinsic relevance of similar regions for regressing region-based saliency to the smooth state. Considering the objects of different size, we exploit the algorithm of multi-scale integration, which generates a more continuous and smooth result. Furthermore, object location is also utilized to suppress background noise for the object-level map, which acts as a vital prior for saliency detection. The final saliency map could be obtained by combing the object-level map with the multi-scale integrated saliency map.The proposed model is evaluated and compared with twenty two state-of-the-art methods on four publicly standard salient object detection databases. Experimental results show that the proposed method comfortably outperforms other state-of-the-art methods in terms of PR curve, F-measure and visual quality.