Efficient Visual Saliency Detection via Multi-Cues

Abstract

Saliency detection has become a valuable tool in computer vision processing, which has been attracting a good deal of attention. Although a lot of research has been done, it cannot obtain ideal performance. In most saliency detection framework, the single difference cue is often used to detect saliency map, but their results were far from satisfaction for low discriminative power of single cue. In addition to the composition of the detection cues during the whole process, the used cues which could only describe the low level information could also lead to poor detection. In order to solve this problem effectively, the paper proposed a novel saliency detection framework by fusing multi-cue difference with high level difference. Basically, the comprehensive information is coupled into multi-cues vectors to remove the non-salient regions and enhance the brightness of the salient area. Specifically, we utilize the fusion of multi-cues and high-level information to improve the ability of understanding images. To further improve the performance of the proposed framework in AUC, we adopt multiple assignments strategy while enhancing the precision of saliency detection. Extensive experiments indicate that the newly constructed multi-cues with high level information could effectively suppress the influence of background information on salient regions. In order to verify the effectiveness of the new algorithm, our experiment uses several standard benchmark datasets (MSRA, ASD, SED1, SED2, and SOD) to test the performance of the algorithm. The experimental results demonstrate that this method has achieved good saliency detection result and good AUC performance in the test. More importantly, the experimental results also show that the proposed method is well complementary to many existing algorithms.