Object discovery on RGB-D data via salient object proposals

This paper presents an effective approach for object discovery in which both object proposals and saliency information are exploited. Our algorithm consists of three basic steps. Firstly, a set of object proposals are generated. Secondly, a saliency map is calculated and salient blobs are detected on the calculated saliency map. Finally, a saliency integrated objectness measure is proposed to rank object proposals thus objects are discovered. Experiments on a dataset of object discovery demonstrate the effectiveness of the proposed method.