Research On Image Reconstruction Algorithm For Sparse Sampling

Tomography is widely used in non-destructive testing,over-the-horizon radar imaging and other fields.Image reconstruction is the core of tomography.As the core of tomography,image reconstruction has been developed for many years.However,with the emergence of new applications and new demands,image reconstruction algorithms face many shortcomings.Among them,the problems of image reconstruction under the background of sparse sampling has attracted extensive attention due to the ill-posed of the solution.On the one hand,new theories such as compressed sensing are introduced.The image feature values are obtained by sparse representation of the image,and then the noise is extracted to approximate the original image during the iterative process.On the other hand,some classical image reconstruction algorithms are combined with artificial intelligence methods such as neural networks,which attempt to establish a mapping relationship between under-sampled reconstructed images and fully-sampled images.However,it is difficult to recover projection data and reconstruction due to under-sampling.Aiming these issues,the main research work and innovation points of this paper can be summarized as follows:(1)A PGS-MTGV image reconstruction algorithm based on non-local prior constraints is proposed.For the total generalized variation model cannot fully utilize the self-similarity information of the image structure,an improved generalized variation image reconstruction model under non-local constraints is established to improve the quality of image reconstruction in the sparse sampling situation.This model introduces a non-local self-similarity of the transform domain as a priori information for image reconstruction.And the multi-directional total generalized variation regularization constraint is calculated in the eight-neighborhood space.Secondly,this paper uses augmented Lagrange theory to remove the constraint,and the model solving method based on improved generalized total variation is derived.Finally,an image reconstruction algorithm is further proposed based on the above research.In the iteration,the algorithm first corrects the prior image according to the non-local information,and then solves the improved model to obtain the reconstructed image.The experimental results show that the proposed PGS-MTGV algorithm can effectively remove artifacts in the image and meet the requirements of image reconstruction quality under sparse sampling.(2)A AWAT reconstruction algorithm based on hybrid regularization model is proposed.Firstly,aiming at the problem that TV regularized algorithm is easy to be sensitive to noise and artifacts in under-sampling environment,a dual regularization adaptive weighted image reconstruction model based on discrete wavelet and TV is constructed.Based on this model,an adaptive weighted iterative reconstruction algorithm is proposed.In each iteration,the algorithm first calculates the TV regularization terms and the wavelet coefficient prior terms by the threshold shrinkagemethod,and then updates the image to be reconstructed.At the same time,in order to further improve the quality of the reconstructed image,and the iterative support set detection method is introduced to calculate the adaptive weight of the wavelet coefficient,which reduces the requirement of the reconstruction algorithm for the projection data size.The experimental results show that the proposed algorithm can achieve better overall performance than LSQR-STF and L1 TV algorithms in terms of time efficiency and reconstruction quality.