Research On Structured Measurement Matrix And High Performance Reconstruction Algorithm In Image Compressive Sensing

The size of the actual image is large,and the hardware system is under great pressure in the huge image information acquisition and transmission process.The traditional image compression method is based on the Nyquist principle.It requires that the sampling rate should not be less than twice the signal bandwidth to ensure that the original signal can be accurately reconstructed.The mode of traditional image compression method is that the signal is usually sampled firstly and then compressed,which will lead to a lot of waste of resources,and the excessive sampling rate also cause significant pressure on sampling equipment even though the traditional image compression method can achieve a high compression ratio.The theory of Compressive Sensing breaks the limitations of the Nyquist theory,the image signal can be sampled at a lower sampling rate and the compression process is completed at the same time,which brings a new idea for image compression.Therefore,it is of great significance to study the image compression method based on the compression perception theory according to the actual demand.This thesis makes a targeted research on the structured measurement matrix and the greedy class reconstruction algorithm combined with the actual image compression requirements.The work is as follows:1.In the actual image compression,the measurement matrix must have lower complexity with high sensing performance.In this thesis,an Orthogonal Block Circulant Matrix with block circulant structure is constructed based on the commonly used orthogonal matrix,and the Structured Random Orthogonal Block Circulant Matrix is constructed combining the composition of Structured Random Matrix.The proposed matrix has low complexity and can satisfy the constraints of the measurement matrix with high probability,and also can achieve a fast sampling rate when applied to large-size images.The simulation results show that the quality of the reconstructed image obtained by the proposed matrix is high,which verifies that the proposed matrix has high sensing performance.The proposed Structured Random Orthogonal Block Circulant Matrix has a certain impetus to the development of image Compressive Sensing.2.The greedy algorithms is suitable for image reconstruction with low complexity and fast speed,however the reconstruction accuracy can not meet the actual demand.Combined with the Generalized Orthogonal Matching Pursuit algorithm,this thesis designs Regularized Generalized Orthogonal Matching Pursuit algorithm based on Dice coefficient,which uses the Dice coefficient with excellent performance to select the most matching atom and removes some of the wrong atoms by the regularization method.The algorithm ensures that the atoms are more accurate for the signal estimation at each iteration,so it can reduce the number of iterations and improve the reconstruction accuracy.The simulation experiment results show that the proposed algorithm has high performance compared with the Generalized Orthogonal Matching Pursuit algorithm.