Speaker
Description
Neutron tomography is a crucial tool for material examination, but ring artifacts can significantly decrease data quality and complicate tasks like segmentation and morphological analysis. The Block-Matching and 3D filtering (BM3D) algorithm, known for mitigating vertical streaks in sinograms and addressing the root cause of ring artifacts, is unfortunately slow and CPU-intensive. We introduce a unique, open-source software solution that eliminates ring artifacts in neutron tomography using the BM3D algorithm. By leveraging both CPU acceleration through Numba and GPU acceleration through CuPy, our approach significantly improves computational efficiency while maintaining data integrity. This dual-acceleration framework drastically speeds up BM3D processing, allowing researchers to quickly obtain refined results and streamline segmentation and morphological analysis.
| Abstract publication | I agree that the abstract will be published on the web site |
|---|
