Leveraging global and background contrast for salient object detection in hyperspectral images

Salient object detection in hyperspectral images has attracted increasing attention, thanks to the rich spectral information that is beneficial to distinguishing salient objects from cluttered background. Most existing methods adapt the color components of Itti's model to spectral domain to detect salient objects. However, these methods typically calculate a single saliency map for the foreground, which fails to suppress background clusters effectively. In this paper, we propose a simple yet effective model for salient object detection. Specifically, we first segment a whole hyperspectral image into superpixels. Then, we calculate global and background contrast values for each superpixel based on its spectral features, and also compute a center prior to incorporate spatial location of each superpixel. The final saliency score is obtained through production of the above three values to fully capture spectral-spatial structures of the image. Extensive experiments on real hyperspectral images show that the proposed method outperforms several state-of-the-arts, which well demonstrates its effectiveness.