Object detection in SAR image based on bandlet transform

detection in SAR images is a challenging task as these images are inherently affected with speckle noise. This paper presents a novel algorithm based on bandlet transform for object detection in Synthetic Aperture Radar (SAR) images. Here first a bandlet based despeckling scheme is employed on the input SAR image and then a constant false alarm rate (CFAR) detector is used for object detection. The input image is first decomposed using Bandlet transform and the bandlet coefficients so obtained are modified using soft thresholding rule on all sub bands, except for low frequency sub band. The optimum thresholds for each sub bands are computed using generalized cross-validation (GCV) technique which doesn't require the information on noise variance of the input image. The method takes advantage of the geometrical features of bandlet transform for retaining the edges and boundaries of the objects, present in SAR images while removing the speckle effectively. Thus CFAR detection on despeckled image can effectively find an optimum threshold for object detection to maintain a constant false alarm rate. The proposed Bandlet transform based scheme surpasses the traditional despeckling and object detection schemes in wavelet domain, in terms of numerical and visual quality. (C) 2016 Elsevier Inc. All rights reserved.