Low Sampling Rate Reconstruction of Medical Imaging: Application of Targeted Sampling Based on OMP

Abstract

Recent theory of compressed sensing informs us that near-exact recovery of an unknown sparse signal is possible from a very limited number of wavelet samples by solving optimization problems. The significance of compressed sensing theory is not only to make much fuller use of recent limited resource of bandwidth, but to break the traditional sampling model which contents sampling, compressing, transferring, decompressing, leaving the data processing part(decompressing) which is much more difficult to computer terminal with higher computational capabilities. The advantage is that we can solve many problems or strengthen local function in the new system model. In the application of medical imaging, less sampling means less time and less harm, which is a great meaning to patients. This thesis is mainly aimed at an relatively mature algorithm OMP(Orthogonal Matching Pursuit) on the reconstructing to different class or size of images, to analyze and solve the problems in the reconstruction. In the experimental process, for the problems that large luminance difference in some part results in the inferior reconstruction, we propose to improve the reconstruction of the part we are interested by up sampling, while down sampling the rest. By sampling targetedly based on OMP, we improved the PSNR of the reconstruction with no more samples to the whole image. In consideration of the characteristic of medical image that the information is relatively concentrated, the method can be operable and practical in real applications.