DescriptionThis work aims at developing high quality and robust 3D reconstructions from X-ray 2D and 3D cone beam CT scanners in the presence of noise and using sparse views. - We utilize the flexible numerical optimization framework of proximal algorithms like the ADMM method.
- We developed a high quality solver based on the SART algorithm and combined it with several regularizers that achieved better results that state-of-the-art.
- We extended a structure tensor-based regularizer to 3D and applied it to specific structures such as thin sheets and fibers.
- We also developed solvers for different noise models, and advanced structure tensor-based regularizers for handling special structures in the scanned objects.
- We compared different proximal solvers for the tomography problem, including SART, ART, BICAV, and Ordered Subsets.
- We developed solvers for the Gaussian and Poisson noises.
- We performed extensive experiments on 2D synthetic and real datasets.
Software- TRex toolbox for X-Ray CT Reconstruction.
References- Guangming Zang, Mohamed Aly, Peter Wonka, and Wolfgang Heidrich. Super-resolution CT Reconstruction. In preparation, 2017.
- Mohamed Aly, Guangming Zang, Wolfgang Heidrich, and Peter Wonka. TRex: A Tomography Reconstruction Proximal Framework for Robust Sparse View X-Ray Applications. arXiv preprint arXiv:1606.03601, 2016, [pdf]
This is joint work with Guangming Zang, Wolfgang Heidrich, and Peter Wonka.
|
|