RGBD Image Co-saliency Object Detection Based on Sample Selection

Co-saliency object detection aims to discover common and salient objects in an image group which contains two or more relevant images. In this paper, a method of using machine learning is proposed to detect co-saliency objects. Firstly, some simple images are selected to form a simple image set based on four scoring indicators. Secondly, positive and negative samples are extracted from the simple images set based on co-coherence characteristics, and high-dimensional semantic features are extracted by the deep learning model which receives RGBD four-channels input. Thirdly, the co-saliency classifier is trained by positive and negative samples, and co-saliency maps are generated by testing all the superpixels in the images by the co-saliency classifier. Finally, a smooth fusion operation is adopted to generate the final co-saliency map. Experimental results on the public benchmark dataset show that the proposed algorithm is superior to the state-of-the-art methods in terms of accuracy and efficiency, and it is robust.