Building Detection in Very High Resolution SAR Images Via Sparse Representation Over Learned Dictionaries

The spatial resolution of synthetic aperture radar (SAR) images has significantly increased in the past few years. As a result, very high resolution images are available for interpretation of the urban area. Buildings are one of the most important parts of the urban area. In this paper, we propose a novel method for building detection from SAR images. First, a set of primary detections, using double bounce, layover, and shadow features, are made. These features are extracted using the order statistics constant false alarm rate method and power ratio detector. The detectors parameters are adjusted in such a way that all bright and dark areas are well extracted. Second, the K-singular value decomposition method is used to learn two dictionaries for discrimination of buildings from clutters. The main idea is to utilize recognition techniques in detection algorithms in order to reduce the number of the false alarms. Gray level statistical, textural, and shape features are used to form the dictionaries. In this step, a secondary set of detections are available, and the final detections are made by the fusion of these two sets. The effectiveness of the algorithm is evaluated by two different TerraSAR-X images from a densely built-up area. The results show that the proposed method is an effective way to detect building through single SAR images.