The Research Of Image Reconstruction Algorithm Based On Compressed Sensing

In the field of modern signal processing, the rapid development of information technology makes the data grow rapidly. The sampling method guided by the Nyquist sampling theory makes the amount of data increasing. So hardware system is hard to meet the requirements of actual needs. In 2006, Candes etal proposed the theory of compressed sensing (CS), which breaks the constraint of the traditional Nyquist sampling theory. A measurement matrix is designed to finish sparse measurement for signals, and then the image is reconstructed according to the reconstruction algorithm. CS theory samples and compresses data at the same time, so the data reconstruction with low sampling rate is achieved. The following is the main work and innovation of this paper:1. This paper mainly research on the two key technologies:measurement matrix and reconstruction algorithm. At first, we deeply studied and analyzed on measurement matrix. Simulation experiments were done for traditional measurement matrix. The experimental results show that measurement matrix improved by the singular value own higher reconstruction accuracy than the original matrix, and can increase 1-2dB of PSNR with the same running time.2. Several compressed sensing reconstruction experiments are conducted by each algorithm. The block-based image compression perception reconstruction algorithm is deeply studied. Block compressed sensing has the advantages of speed, accounting for small memory, which is prone to image block effects. On this basis, this paper proposes a new compressed sensing image reconstruction algorithm based on region segmentation. Experimental results show that the improved reconstruction algorithm significantly improves the reconstruction effect while decrease the running time of reconstruction.3. Traditional compressed sensing image reconstruction algorithms need measure all the coefficients of multi-wavelet transform to ensure image quality, and the wavelet capture limited direction information. So the reconstructed image quality is poor. Non-subsample shearlet transform is used as signal sparse transform, and according to the characteristics of transform coefficients, coefficients are measured selectively. Experimental results show that only a single layer decomposing of Non-Subsampled shear wave transform can construct high quality image, overcoming the disadvantages of the conventional algorithm in need of a multi-layer wavelet transform, which reduces the amount of data sampling and reconstruction quality of images is extremely improved.