A Global-to-Local Thresholding Method for Object Enhancement in SAR Images

Speckle effect exhibited in Synthetic Aperture Radar (SAR) images significantly influences the interpretation performance of the follow-up tasks, such as object recognition. Although many contrast enhancement methods have been proposed for discriminating objects from the background, their performances often degrade in SAR images. To achieve better contrast between objects and the background, a global-to-local thresholding method is proposed for SAR object enhancement. Specifically, we first obtain two global thresholds for object segmentation based on entropy maximization and Ostu methods. Then, detailed object structures can be enhanced based on an adaptive local thresholding. After several iterations of the local processing, we can obtain the enhanced SAR objects with high contrast with respect to the background. Based on the experiments on real SAR images, the proposed method demonstrates better performance than other contrast enhancement algorithms.