Saliency detection via extreme learning machine

In this paper, we propose an effective algorithm based on Extreme Learning Machine (ELM) for salient object detection. First, saliency maps generated by existing methods are taken as prior maps, from which training samples are collected for an ELM classifier. Second, the ELM classifier is learned to detect the salient regions, and the final results are generated by fusing multi-scale saliency maps. This ELM-based model can improve the performance of different state-of-the-art methods to a large degree. Furthermore, we present an integration mechanism to take advantages of superiorities of multiple saliency maps. Extensive experiments on five datasets demonstrate that our method performs well and the significant improvement can be achieved when applying our model to existing saliency approaches.