Measurement Matrix Design And Reconstruction Algorithm Of Image Compressed Sensing

Compressed sensing theory is a new signal processing technology which is widely used in the field of image processing, and it uses a small amount of sampling signal to accurately recover the original signal. Therefore, based on the analysis of the existing measurement matrix and reconstruction algorithm, this paper carries out a thorough study of the measurement matrix optimization and reconstruction algorithm.Firstly, in order to reduce the correlation between the sparse matrix and the measurement matrix, this paper put forwards an optimization approach for improved measurement matrix, using a gradient descent method and projection approximation method to optimize the measurement matrix. At the same time, the optimal step size is carried out adaptively processing, the amounts of iterations of the measurement matrix optimization are reduced. The experimental results show that the optimization of the measurement matrix can not only improve the quality of the reconstructed image, but also can effectively reduce the time of the optimization of the measurement matrix.Secondly, aim at solving the problem of inaccurate noise variance estimation in the image reconstruction algorithm with fixed noise scale factor, this paper presents a noise scale factor which is based on image texture feature, sparse transform type and sampling rate of joint estimation. Using the noise scale factor to substitute fixed noise scale factor and improve noise variance estimation can not only to accurately estimate the noise variance in the image, but also conducive to the removal of noise in the reconstruction algorithm. The experimental results show that the improved noise variance estimation is benefit to the image reconstruction algorithm of removing the noise and can improve the quality of the reconstruction.Finally, based on the defects of using the variance of single wavelet coefficient as threshold parameters in the bivariate threshold shrinkage denoising model image reconstruction algorithm, the variances of the wavelet coefficients as a threshold parameter, the father wavelet coefficients and neighborhood wavelet coefficients were used as threshold parameters in this paper to construct variables wavelet threshold shrinkage model to improve the accuracy of threshold estimation. Meanwhile, the values of the current wavelet coefficients, the father wavelet coefficients and the neighborhood wavelet coefficient were judged after model processed to eliminate the excess noise in the image.Experimental results show that the improved wavelet variable threshold shrinkage model can further reduce the noise in the image and improve the quality of the restored image.