Object-based 3D X-ray image reconstruction

The representation of a complex scene as an assembly of discrete 3D objects is of value in many industrial, medical, and security applications. Computed tomography (CT) entails very high X-ray exposures and a heavy computational load. It produces excellent pixel-based 2D images, but the third dimension depends on interpolation between some tens of slices. No-one attempting a 3D equivalent of CT has found an acceptable combination of resolution, signal/noise ratio, and computational load. Since we are not concerned with millions of independent voxels, but with a few discrete multi-voxel objects, some authors have used parallel projection of object outlines, from say 10 distributed views, to define an outer bound to the surfaces of these objects. The resulting representation is inaccurate, and topologically limited, but the radiation is reduced about 1000-fold. Our method, nicknamed object-3D, extracts from the same information the precise 3D structure of the objects