Study On Satellite Image Restoration Based On Sparse Representation And Adaptive Reciprocal Cell

Satellite image plays an irreplaceable role in aerospace, national defense and military affairs applications. However, many steps exist during image acquisition will strongly influence the quality of observation results, such as remote sensors, sampling system and the signal transmission. This may restrict the quantitative analysis of image and subsequent processing. Therefore, improve the quality of image has become a top topic in remote sensing field.In this paper, based on the exploration of satellite imaging system, we mainly focus on studying accuracy and effective methods to meet practicability. In practical application, we have respectively studied methods of sparse representation and adaptive reciprocal cell (AR-cell). Sparse representation can effectively obtain essential features of an image using linear combination of a few dictionary atoms. It provides a new perspective for image processing.Adaptive reciprocal cell restoration is based on satellite image. This method constraints frequency spectrum reasonably, and then combines with conventional restoration to denoise and deblur. Our work includes the following aspects(1) The nonlocal sparse representation model and algorithm for image denosing are analyzed intensively. First, we present a nonlocal regularization term using the similarity of image structure, and introduce it into traditional sparse representation model. Then, we employ majorization method and substitute the original objective function with a surrogation function, proposing sparse representation and dictionary learning using the majorization minization method based nonlocal frame. This method can guarantee to find local minima in each optimization step. We use standard test images and satellite images validate the method.(2) We analyze the AR-cell model and develop AR-cell bank which adapt to similar image structures. Then we introduce the gradient-fitting term and nonlocal operator into the total variance model, proposing adaptive reciprocal cell bank based nonlocal restoration method. In addition, a splitting method is considered to solve the model. Experiment results show the promising performance of the proposed method.