Section of Life Sciences and Technologies, EPFL
This project aimed at correcting phase wrapping artifacts on images acquired by Differential Phase Contrast Tomography (DPCT). Phase contrast, in comparison to attenuation contrast, improves soft tissue imaging, but is subject to phase wrapping. Phase unwrapping is an ill-posed problem and there is no general way of retrieving the best solution. Several different approaches were designed and compared to perform phase unwrapping of the sinograms obtained by DPCT. Some direct methods were implemented as well as some combinatorial and iterative techniques. Direct methods are generally simpler and faster to compute, but are error-prone, while indirect methods show a higher complexity and possess several parameters that have to be optimized, but are then more robust. As long as it remains computable, the combinatorial approach should be the most powerful technique, since it allows a wide variety of cost functions and does not suffer convergence problems. All techniques managed to correct the main wrapped regions of the provided sinogram, but none of them was able to correct all artifacts. The designed methods allow to find solutions optimal in some sense. It seems that a different optimality criterion should be defined in order to correct the remaining artifacts.