Intra-Saliency Transfer for Effective Salient Object Detection

Abstract

A new color contrast based, bottom up saliency method called INS has been proposed in this paper. Intra-saliency transfer and multiple distance functions have been exploited effectively in the INS method. A color-based super-pixel segmentation approach has been used in the INS method in order to decompose the image into regions. Further, three distance functions are used to measure the saliency to effectively characterize the salient regions. These functions are evaluated and integrated to effectively determine the final saliency map. The INS method has been evaluated exhaustively on the MSRA-1000 benchmark image dataset available in public domain. Further, the results of the INS method have been compared with those of twelve state-of-the-art salient object detection methods found in recent literature. The INS method has exhibited superior saliency detection performance compared to the methods used for comparison.