| Image acquisition is a process of collecting the nature signal into the computer by using some kinds of sensors. With the development of image in all aspects of application, the high image resolution, faster, and the wider-band imaging needs, as well as expensive detection units and the limited of image acquisition side of energy, have made that it necessary to study an image acquisition framework which reduce the burden in image acquisition. The compressive sensing as a new concept for signal acquisition provides a new way, by measuring the signal that satisfy RIP conditions, and by using the measurements the method can reconstruct the original image. This study is based on above background and by studying two key parts of the compressive sensing imaging system - measuring and reconstruction, then research on the condition of CS can be used. Through analysis the classical algorithm, the thesis proposed effective improvement for CS imaging system and improve the performance of the system.Firstly, this thesis research on the CS performance and the sparsity for different kinds of images. The results of image sparsity showed that, astronomy and MRI image, have a good sparsity, and the sparsity of panchromatic remote sensing image is worst. By analyzed the relationship between CS performance of these four images and the sparsity, the research get a conclusion that the application of compressed sensing is suitable for several types of sparse or compressible images. Finally, through analyzed CS information acquisition, the results show that the CS acquisition process maintains spatial structure for the original signal.Secondly, because of the choice of reconstruction methods influence the number of measurements and the computational complexity of image acquisition system, the part of CS acquisition has been studied. In this thesis, three typical methods have been studied: (1) the perception method based on random matrix (2) perception matrix based on transformation (3) structure random perception method. The thesis analyzed the performance of three methods for the acquisition, the hardware application, computing and storage consumption. The perception random matrix can be used in widely range of applications and hardware is convenience to realized, but these can not adjust the projection matrix about the image sparsity to achieve more optimal projection. Based on above research, the thesis proposed a projection correlation minimization method, which search a matrix has a low relationship with some sparse basis. Experimental results show that the projection correlation minimization method was better than the random matrix and the peak signal to noise ratio improved.Finally, the thesis studied the method of CS reconstruction. The CS measurements need to be reconstructed in order to be used in following application and the reconstruction algorithm usually has a lot of iteration operation in the system reconstruction. The CS reconstruction algorithm have a significant impact on CS performance, the speed of system and accuracy of the CS imaging system, so two kinds of typical reconstruction algorithms have been studied: a reconstruction algorithm based on greedy algorithms and the convex optimization reconstruction algorithm based on gradient. The typical reconstruction method can achieve a certain performance in both speed and accuracy, but these methods which based on the l 1 norm minimization reconstruction algorithm to approximate l 0 norm reconstruction (unsolvable), are not precise enough to reconstruct the large coefficients, so the performance of the recovered image can be improved further. Based on the above research, this thesis proposed an iterative weighing algorithm for image reconstruction in CS. The experimental results show that the reconstructed image performance has been significantly improved, and the image results can converge in a relatively short time. Compared with traditional reconstruction methods, the proposed method is available in CS imaging system. |
PDF Full Text Request