A bottom-up saliency-based segmentation for high-resolution satellite images

Abstract

High-resolution satellite images, especially of urban areas, are complex in nature. Presence of multiple objects renders analysis of such images challenging. As human mind is best in dealing with complex images and ambiguous problems, such urban area satellite images can be best interpreted by development of such systems that can emulate human mind. Therefore, the presented work separates the high-resolution satellite images into foreground and background on the basis of saliency for simulating human-like segmentation. This work exploits the existing saliency methods available along with segmentation algorithm. Results have been analyzed on the basis of the robustness of ‘shape’ which is a key factor for object recognition. Presented work also propose a measure to assess the goodness of the segmented images which is based on similar segmentation of terrestrial images in a separate work. © Springer Nature Singapore Pte Ltd. 2018.