Computed Tomography (CT) is a powerful, nondestructive technique for producing 2-D and 3-D cross-sectional images of an object from X-ray images. Conventional CT requires the object to be inserted into the CT scanner and that many X-ray images are taken, prior to reconstruction of the 3D image. However, there are many situations in which these requirements are not met: If the object is a fixed part of an object that is too large to fit in the scanner object; if moving the object is dangerous (e.g., explosives), if moving the object would disrupt or pollute the context (e.g., crime scene), if the object cannot easily be transported for X-ray scanning (e.g., horse with broken bone), or if the object is too valuable to be removed (e.g., cultural heritage).
To cope with such situations, in situ X-ray scanning is required. Portable X-ray devices, such as a hand-held X-ray camera or a robot system, are available on the market, which are, however, intended to acquire only a single or a series of X-ray images. It is currently not possible to produce a 3D reconstruction of the object from these X-ray images. This is because:
1. in contrast to common CT-scanners, the exact position and orientation of the X-ray source/detector system with respect to the object is unknown.
2. if a hand-held camera or robot system is employed, the scanning process is time consuming, which limits the number of X-ray images to be acquired for tomography. Current CT reconstruction algorithms require a large number of X-ray images to obtain accurate results.
3. it may not be possible to acquire projection images from all angles. Both 2 and 3 result in a highly underdetermined inverse problem.
This project aims at the development of robust, efficient reconstruction methods for in situ X-ray scanning & tomography. These methods will not require accurate prior knowledge of the scanning geometry, and will be tailored for achieving maximal reconstruction quality from a small number of projection images. To this end, the following computational strategies will be combined:
1. Automatic parameter estimation based on consistency maximization of the simulated X-ray images with respect to the measured X-ray images will allow the reconstruction algorithm to deal with unknown geometrical parameters.
2. New algorithms will be developed for efficient optimization of the high dimensional search space (including the unknown object volume and the position/orientation of the acquired X-ray images) by exploiting the linearity of certain reconstruction algorithms and exploring multi-resolution approaches for gradual refinement of parameter estimates.
3. Compressive sensing and discrete tomography will be incorporated within the parameter estimation framework to allow for accurate image reconstruction from few projections and deal effectively with a limited angular range.
4. State-of-the-art GPU computing techniques, based on recent advances in a current research project on accelerating tomography algorithms, will be employed to effectively deal with the high computational requirements.
A successful project will open up numerous applications in various fields such as security, nondestructive testing, dental imaging, veterinary imaging, or cultural heritage.