Salient object detection via multiple saliency weights

摘要

Salient object detection aims to emulate the extraordinary capability of human visual system, which has the ability to find the most visually attractive objects in a complex visual scene. The human visual attention is often complicated and affected by many factors. In this paper, we present a novel bottom-up approach to automatically detect salient objects of an image via multiple visual cues. The key idea of our approach is to represent a saliency map of an image as an integration of multiple visual cues (saliency weights), which have been proven to be effective and useful. Specifically, we propose four saliency weights, i.e., local contrast weight, superpixel clarity weight, background probability weight, and central bias weight, to effectively represent each visual cue. To obtain our saliency map, the four resulting saliency weights are integrated in a principled way via multiplication and summation based fusion. Furthermore, we propose a new superpixel-level saliency smoothing approach to optimize the integrated results for producing clean and consistent saliency maps. Our experimental results on three standard benchmark datasets demonstrate that the proposed approach outperforms other saliency detection approaches in terms of the subjective observations and objective evaluations.