Saliency detection based on Boosting learning

In this paper, we propose a method for saliency detection based on Boosting algorithms in still images. Compared to saliency detectors of pixel level based, we detect salient regions of an image based on sub-windows at any locations and sizes. For each window, we compute a set of features including local contrast, gradient histogram contrast. We construct our detector based on a cascade AdaBoost classifier to get the sub-windows which contain salient objects. Generally, more than one sub-window would get through the AdaBoost detector and we introduce a score function to remove redundant sub-windows and get the final one. The algorithm is tested on the MSRA Salient Object Database, and experiment results show that the proposed approach achieves a fast and accurate saliency detection system.