Research On Saliency Detection Method In SAR Images

Human vision system, with efficient image interpretation ability, could detect salient region and extract interesting targets rapidly. This thesis aims to improve SAR image interpretation ability by using visual attention mechanism theory, and propose a saliency detection algorithm for SAR images. In this paper, the typical saliency detection algorithms have been summarized and analyzed in detail. And then this thesis, considering of the features of SAR images, puts forward a target detection algorithm and scale self-adaptive saliency detection algorithm for SAR images. The main work includes the following aspects.(1) The typical saliency detection algorithms have been summarized and analyzed. At first, feature fusion theory and visual attention frame based on neurobiology have been analyzed in detail. Besides, objective evaluation criteria of saliency detection algorithms have been discussed. Furthermore, the typical saliency detection algorithms have been summarized and analyzed explicitly. Finally, real data of optical and SAR images are used to evaluate the detection performance of the above four algorithms.(2) A target detection algorithm based on visual saliency for SAR images has been proposed. To begin with, this thesis addresses with the statistical modeling of background clutter in SAR images, summarizes typical statistical models used for background clutter and also enumerates several optimal statistical model selection criteria. Then, dual parameter CFAR detector based on statistical model has been discussed. After that, a target detection algorithm for SAR images has been proposed by combining visual saliency theory and CFAR detector. At last, the comparison of the detection results shows that the proposed algorithm detects all size-fixed targets with lower false alarm rate than dual parameter CFAR detector.(3) Scale self-adaptive saliency detection algorithm for SAR images has been proposed. Based on salient region detection algorithm for optical images proposed by Kadir, scale self-adaptive saliency detection algorithm for SAR images, considering of the difference between optical images and SAR images, has been proposed. First of all, the local complexity metric and self-dissimilarity metric have been redefined to be more suitable for SAR images. Then saliency scale determination method has been improved, which is more accurate. Finally, the saliency map has been built by combining saliency metric with saliency scale, which is the last step of saliency detection. Experiment results show that the proposed method is more reliable for scene analysis of SAR images than Kadir's algorithm, and that the proposed method is more effectively for saliency detection of SAR images than aforementioned typical saliency detection algorithms.