LAUSR

Abstract Iterative reconstruction algorithms are often needed to help solve ill-posed inverse problems in computed tomography (CT), especially cases when tomographic projection data are corrupt, noisy or angularly undersampled. Model-based iterative methods can be adapted to fit the measurement characteristics of the data (e.g. noise statistics) and expectations regarding the reconstructed object (e.g. morphology). The prior information is usually introduced in the form of a regulariser, making the inversion task well-posed. The CCPi-Regularisation toolkit provides a set of variational regularisers (denoisers) which can be embedded in a plug-and-play fashion into proximal splitting methods for image reconstruction. CCPi-RGL comes with algorithms that can satisfy various prior expectations of the reconstructed object, for example being piecewise-constant or piecewise-smooth in nature. The toolkit is written in C language and exploits parallelism with OpenMP directives and the CUDA API; and is wrapped for the Python and MATLAB environments. This paper introduces the toolkit and gives recommendations for selecting a suitable prior model.