Salient object extraction based on nonparametric kernel density estimation

A major problem in content-based image retrieve (CBIR) is how to extract the perceptually salient object in an image. In this paper, we propose an efficient approach for automatic extracting the salient objects. First, an input image is segmented into homogeneous regions based on nonparametric kernel density estimation (NKDE), and then different features representing colour, texture and spatial position for individual region and adjacent region are extracted. By calculating the object important index (Oil), salient objects are adaptively extracted according to the defined criteria. Experimental results demonstrate the excellent extraction performance of the proposed approach.