CT Reconstruction on Unstructured Mesh for Multi-material Object

Computed tomography (CT) is an inspection tool attracting increasing attention in industry for the last decade because of its ability to inspect an object with a non-destructive manner. In general, the boundary is defined as an isosurface of a CT image that is a surface consisting of points with the same CT value. However, a boundary between two materials with different densities tends to incorrectly be located inside of the object with the lower density. In this presentation we propose a novel algorithm extracting such a boundary accurately. In order to suppress a miss location of boundary, we give another definition of boundary based on the directional derivative of CT value. The input of our algorithm is a sinogram, a set of radiographic images projected from various angles. A CT image is computed from a sinogram, so it can be expected that artifacts on a sinogram are less than CT image. Therefore, by taking a sinogram as input, we can compute accurate derivatives and avoid artifacts arising on a CT image. Along with the computation of differentials, we compute a CT image from the sinogram. For this process, we adopt an unstructured mesh, for instance a triangular mesh or a tetrahedral mesh, as an underlying structure on which a CT image is reconstructed, since it has strong advantages over a structured grid (voxels) on a well representation of surface and less number of triangles on the resulting surface mesh. We will also show the effectiveness of the proposed algorithm with some experimental results.