A Study Of SAR Image Despeckling And Image Reconstruction Model Based On Compressed Sensing

The classical approach of reconstructing original signals or images from measured data follows the well-known Shannon sampling theorem, which states that the sampling rate must be at least two times than its highest frequency for accurate reconstruction. Similarly, the fundamental theorem of linear algebra suggests that the number of collected samples of a discrete finite-dimensional signal should be at least as large as its length in order to ensure reconstruction. This theorem underlies most devices of current technology. As perception system is widely used, the amount of data generated by sensing systems has grown from a trickle to a torrent. Which cause measurements limited by physical constraints or extreme expenses.The compressed sensing theory proposed provides a new method for data acquisition. Which subverts traditional wisdom. It predicts that certain signals or images can be reconstructed from what was previously believed to be highly incomplete measurements. Many issues have a similar form with compressed sensing in the image processing; such as image reconstruction, image de-noising, etc., We analyze and deal with that issues from new perspective by studying reconstruction model and algorithm of compressed sensing.This paper is based on application and reconstruction model of compressed sensing theory. We propose: 1. Applying the reconstruction model of compressed sensing to Synthetic Aperture radar(SAR) image de-noising, which may improve SAR de-noising effect. 2. We add structural characteristics of the image to compressed sensing, change image reconstruction model, improve the quality of reconstructed image, And analyze performance of the new models and algorithm. Main work are as follows:(1) Proposing SAR image de-noising model based on sparse reconstruction and Bregman regularization. The multiplicative speckle noise model of the SAR image is converted to the additive noise as inputted image in the sensing system,which derive new SAR de-nosing model. Then, we use spa RSA framework and Laplacede-noising threshold based on Bayesian estimationas to remove speckle from SAR image, and introduce Bregman regularization into de-noising algorithm.Which can accelerate iterative algorithm and enhance the de-noising effect. Experiments show that comparing with traditional algorithm, the new proposed model and the corresponding de-noising algorithm have greatly improved the de-noising effect, preserve more border details and control the runtime. The experimental results show the effectiveness of the new SAR image model.(2) Proposing the compressed sensing model based on texture and cartoon(piecewise smooth)(TC_CS).we propose the new compressed sensing model which have been introduced morphological component analysis(MCA) and cartoon texture model. Which incorporates structural features of the image to avoid over-smoothing and preserve more border and fine structures information. In the TC_CS model, firstly, the image is decomposed into texture and cartoon. Secondly, they are reconstruct by different basis and threshold strategies. Finally, we get the restored image by combining two parts. The experimental results show the effectiveness of the new model and algorithm.