Salient target detection in remote sensing image via cellular automata

In order to detect salient target in remote sensing images effectively and accurately, this paper propose a target segmentation method based on cellular automata which is usually used as a dynamic evolution model. First, we introduce the background based map to obtain saliency map with the help of a widely used superpixel segmentation method named simple linear iterative clustering. Secondly, cellular automata are employed to produce the elementary saliency map. Then enhanced saliency map can be obtained by maximum contrast of image patch method. Adaptive threshold is calculated to segment the enhanced saliency map. Consequently, the salient target detection and segmentation result can be obtained. Experiments on optical remote sensing images and synthetic aperture radar (SAR) images demonstrate that the proposed algorithm outperforms other methods such as K-means, Otsu and region growing method.