Wieckowski, AdamStopp, FabianKäseberg, MarcKeeve, ErwinIvan Viola and Katja Buehler and Timo Ropinski2014-12-162014-12-162014978-3-905674-62-02070-5778https://doi.org/10.2312/vcbm.20141195https://diglib.eg.org/handle/10.2312/vcbm.20141195.163-172Cone-beam computed tomography (CBCT) is an established standard for both, medical and industrial volumetric imaging. To compute a 3D volume, multiple 2D x-ray projection images of an object of interest are acquired from different directions. Using the geometric information about the acquisition geometry of each image, the volume is reconstructed. Incorrect geometric information (misalignments) leads to blurring and other artifacts in the resulting reconstruction. The exact acquisition geometry is commonly calculated by the analysis of a scan of a dedicated calibration body (off-line calibration). Such approach requires high repeat accuracy of the scanner mechanics and cannot account for non-systematic deviations. Current methods allowing for misalignment correction without a dedicated phantom, e.g. by iteratively adapting the geometry to minimize the arising artifacts, were developed to work with planar trajectories. It poses a problem for open CBCT systems driving complex trajectories. Therefore, we propose an enhanced method allowing for misalignment correction for general trajectories. We developed a new quality function and a flexible modeling for misalignments. We successfully applied our method to real datasets acquired along planar and non-planar trajectories. The correction with our approach substantially increases the resulting volume quality.I.4.3 [Image processing and computer vision]EnhancementGeometric correction