Salient Object Extraction Based on Region Saliency Ratio

摘要

This paper proposes an efficient approach to extract salient objects in an image. A scale-invariant saliency map is first constructed based on a multi-resolution feature contrast calculation, meanwhile the image is segmented into homogenous regions using nonparametric kernel density estimation (NKDE). Then the region saliency ratio of each region combination to its complement is calculated in turn. Finally, salient objects are extracted by maximizing the region saliency ratio. Experimental results demonstrate the effectiveness of the proposed approach.