Compressed Sensing Application on non sparse SAR images based on CoSaMP Algorithm

Abstract

We present in this paper a Compressed Sensing (CS) proposal technique tested and evaluated on a Synthetic Aperture Radar (SAR) image. The image signals have to present on a few non-zero values. Many papers were proposed in this field and most of them were tested on sparse simulated images. This paper purpose is to make CS applicable on any kind of data especially in SAR field. In reality, sparse signals doesn’t exist, and finding a sparse representation may be impossible without an adapted basis or a-priori information. This makes CS unusable on real SAR images. The presented method is an alternative way to find any signal sparsity and avoid a long computing time to generate a dictionary. We combine the FFT transform with a recovery algorithm to compute the sparse representation. This method is tested on a non-sparse real values signal and on Flevoland C-band SAR image. 45% of data are sufficient to recover an exploitable image.