Region of interests extraction based on visual saliency for remote sensing image

This paper proposes a novel region of interests extraction method for remote sensing image based on visual saliency analysis as well as the multi-scale saliency fusion mechanism. Firstly, under the theory of multi-scale and multiresolution analysis, the gaussian pyramid representation is generated. Then, various spatial features are exploited to compute saliency maps under different scales. Finally the saliency map could be calculated by the corresponding level fusing as well as the across-scale combination. The comparative experiments validate the effectiveness of the proposed model.