SAR Image Segmentation by Merging Multiple Feature Regions

Abstract

SAR image is widely used in many fields, such as military, agricultural, and environmental, of its distinct characteristics and advantages. As one of the most critical steps in SAR image processing, image segmentation has always been the research focus. However, SAR images often contain complex scenes and have a significant speckle noise resulting in an unsatisfactory segmentation effect. To address the impacts of intricate texture and noise, we used various features to segment the image, such as gray features, texture features, and morphological characteristics. Firstly, the image is initially segmented using superpixels, and the image is in a state of over-segmentation, but the influence of noise will be largely eliminated. Secondly, we use the normalized grayscale co-occurrence matrix to extract the texture features of each superpixel. Whether the two regions belong to the same category can be judged by calculating the difference between the grayscale histogram and the texture features. Moreover, we propose the concept of bracketing coefficient and choose the order of merging. Finally, the merged residual regions are classified using the K-means method. We conduct experiments on a simulated SAR image and a real SAR image. The experimental results show that the segmentation accuracy of the proposed method has reached more than 95, and it has a good segmentation effect.