Research Of Compressed Sensing In Astronomy Images

As a tool to directly acquire astronomical information, Astronomical image has an irreplaceable role in astronomy. But with the Rapid inflation of Astronomical images in quantity and size, current sampling compression processing technology can’t meet the demand of scientific research. Based on compressed sensing theory and feature of astronomy image, this article carries on deepgoing research on application of compressed sensing in astronomical images and makes improvement to some shortcoming. The main points are as follows:Combining characteristics of astronomical image, this article builds a frame of processing astronomical image with compressed sensing. Daubechies orthogonal wavelet is applied to represent sparsity of images. Then Random Fourier ensemble matrices is designed to measure sparse representation. For the method of reconstruction, we prefer to use the total variation minimization (TV-min) and the results are robust.While the total variation minimization can provide more accurate reconstruction results, it also makes computation speed lower. So it is hard to be applied directly to large scale astronomical image. We combined block compressed sensing proposed by Gan with above frame and the experiments show that the speed of reconstruction was improved more than three times. However, in block CS, large image is divided into small blocks with same size and sampled in a block by block manner with same measurement operator. As a consequence, it can result in a waste of resources. Therefore this article propose Block Adaptive Sampling algorithm(BAS), whose emphasis is that for the sparsity of different blocks, it is sampled with different sensing matrix. And experiments show that BAS also can improve the accuracy of image reconstruction.In order to solve edge block effect of reconstructed image caused by block compressed sensing and according to total variation denoising model proposed by Rubin, we can adjust reconstruction model of the total variation minimization by setting weighted adaptive parameter P(m,n)1<P(m, n)<2).this Parameter is related with pixel. When near edge points, the model can better keep edge features; when pixel far away edge points, it can maintain the accuracy of reconstruction. Through experiments, it proved that this method can reduce the blocking artifact and smooth reconstruct image.