Abstract:Total-variation (TV)-based Computed Tomography (CT) image reconstruction hasshown experimentally to be capable of producing accurate reconstructions fromsparse-view data. In particular TV-based reconstruction is very well suited forimages with piecewise nearly constant regions. Computationally, however,TV-based reconstruction is much more demanding, especially for 3D imaging, andthe reconstruction from clinical data sets is far from being close toreal-time. This is undesirable from a clinical perspective, and thus there isan incentive to accelerate the solution of the underlying optimization problem.The TV reconstruction can in principle be found by any optimization method, butin practice the large-scale systems arising in CT image reconstruction precludethe use of memory-demanding methods such as Newton's method. The simplegradient method has much lower memory requirements, but exhibits slowconvergence. In the present work we consider the use of two acceleratedgradient-based methods, GPBB and UPN, for reducing the number of gradientmethod iterations needed to achieve a high-accuracy TV solution in CT imagereconstruction. The former incorporates several heuristics from theoptimization literature such as Barzilai-Borwein (BB) step size selection andnonmonotone line search. The latter uses a cleverly chosen sequence ofauxiliary points to achieve a better convergence rate. The methods are memoryefficient and equipped with a stopping criterion to ensure that the TVreconstruction has indeed been found. An implementation of the methods (in Cwith interface to Matlab) is available for download fromhttp://www2.imm.dtu.dk/ pch/TVReg/. We compare the proposed methods with thestandard gradient method, applied to a 3D test problem with synthetic few-viewdata. We find experimentally that for realistic parameters the proposed methodssignificantly outperform the gradient method.

Compressive total variation for image reconstruction and restoration

Efficient Compressed Sensing Reconstruction Using Group Sparse Total Variation Regularization

Remote Sensing Image Reconstruction Method Based on Non-Local Similarity and Low Rank Matrix

Dynamic MRI reconstruction based on tensor-SVD

A nonlinear total variation based computed tomography (CT) image reconstruction method using gradient reinforcement

An Efficient Non-Convex Total Variation Approach for Image Deblurring and Denoising

Reweighted Anisotropic Total Variation Minimization for Limited-Angle CT Reconstruction

High-Accuracy Total Variation for Compressed Video Sensing

Image Reconstruction from Incomplete Convolution Data Via Total Variation Regularization

Nonlocal Gradient Sparsity Regularization for Image Restoration

A New Alternating Minimization Algorithm for Total Variation Image Reconstruction

Total variation image reconstruction algorithm based on non-convex function

Image Restoration by Combined Order Regularization with Optimal Spatial Adaptation

Template-based CT Reconstruction with Optimal Transport and Total Generalized Variation

Total Generalized Variation Restoration with Non-Quadratic Fidelity

Sparse-view x-ray CT reconstruction via total generalized variation regularization

Constrained Higher Degree Total P-Variation Minimization for MRI Reconstruction from Undersampled K-Space Data

Improved total variation based image compressive sensing recovery by nonlocal regularization

A Fast Algorithm for Total Variation Image Reconstruction from Random Projections

Tensor Compressed Video Sensing Reconstruction by Combination of Fractional-Order Total Variation and Sparsifying Transform

Accelerated Gradient Methods for Total-Variation-based CT Image Reconstruction