SAR Image Based On Compressed Sensing Bandelets Transform Algorithm

Synthetic aperture radar (SAR) has important application in civilian and military, because ofits all-weather, all-time, long-distance observation abilities. SAR image contains huge of data, inorder to save storage space and transmission time, SAR image compression has become animportant research topic.Compressed sensing can accurately reconstruction a sparse signal by using just of its smallnumber measurement samples, with high probability. Based on compressed sensing, it is canobtain the compressed measurement samples by projecting the SAR image onto a measure matrix,which is irrelevant with the transform basis. In the decompression phase, the original SAR imageis accurately reconstructed from the measurement samples by solving an optimization problem.Compressed sensing is an effective SAR image compression method, but the SAR imageedge information can not be well preserved by the algorithm which uses traditional wavelet basisas the sparse basis.Bandelets transform can adaptively track the geometrically regular direction,effectively sparse represent natural image, and preserve the edge information well during theimage compression. Therefore introduction of Bandelets transform to the SAR images compressedsensing, will be able to effectively maintain the edge information in the SAR image.The paper will focus on the following parts:This article will introduce Bandelets transform into ordinary optical image compression, weobtain the Bandelets basis and then reconstructe the date.We improve the algorithm after the introduction of the second generation Bandeletstransform, sampling is using the random matrix to get the measurement to the original signal. Weuse the rotation matrix to introduce the best geometry information and then we build a new sparsegroup with the introduced information and the Harr wavelet basis. After all, reconstructe the data.We bring the measurement forward to the original signal.We introducte the method to the SAR image compression, and validate that Bandeletstransform of our method has better edge retention than wavelets transform through the simulationexperiment.