To achieve good image quality for computed tomography scans, it is important to accurately know the geometrical relationship between the x-ray source, detector, and axis of rotation. Conventional geometric calibration algorithms generally require particular calibration phantoms, such as a small pin or wire, which may not be practical for all scanner types, particularly for large industrial scanners. This paper presents an alternative framework to calibrate system geometry from scans of arbitrary objects, without prior knowledge of the object's form. This lessens physical construction requirements, and permits post-scan geometric calibration from any arbitrary scan data. Experimental results show that central ray calibration using this approach can give results accurate to 0.01 channels for low-noise conditions, or 0.1-0.45 channels under higher noise levels.
Kevin M. Holt