Region Based Multi-spectral Saliency Detection

Visual saliency detection is useful for applications like image segmentation, image retrieval et al. According to the thought that the near-infrared image can provide complementary information to the corresponding visible image, a new region based multispectral saliency detection method is proposed. The source images are decomposed into homogeneous regions by superpixel algorithm in 6-dimentional vector space which is composed of the LAB colour values of the visible image, the grey value of the near-infrared image and the two-dimensional pixel coordinate. Then the final saliency map is obtained by measuring the region’s contrast which is computed by using multi-spectral colour information and spatial distance. The publicly RGB-NIR dataset is adopted for experiment evaluation; Compared with five other state-of-the-art methods, the proposed method produces superior performance. Experimental results show that the proposed method is effective.