Elliptical region of interest based saliency detection

Abstract

The two main tasks of computer vision problems are extraction of image descriptors and their subsiquent matching. The choice of type and level of description is task dependent, which is a low-level approach for solving computer vision problems. Saliency detection is a basic task in perceptual organization. Without any prior assumption, automatic estimation of salient object regions in an image enhances the computer vision applications. In this paper, we proposed a novel saliency detection algorithm, which evaluates the local saliencies over random elliptical regions of interests. Though the proposed algorithm is multi-scale approach, it is simple, efficient, and gives high quality saliency maps with higher resolution as compared to other existing methods. The qualitative and quantitative results of the proposed saliency deaction algorithm consistently outperforms other state-of-the atrs realizing higher precision with better recall rates and F-measure value.