3D tomographic reconstruction of binary images from cone beam projections: a fast level set approach

This paper addresses 3D tomographic reconstruction of a binary image from a small number of cone-beam projections. Under the binary hypothesis, the object may be conveniently represented by its surfaces, and the reconstruction problem solved using active contour methods. To handle complex topologies, we formalize the problem in the framework of level sets using implicit descriptions of object boundaries. The expression of a velocity including projection data is derived in this context. In addition, regularization is introduced in order to manage both lack of data and noise. A fast implementation of the level set method is described. Results are presented both from realistic simulations and physical data.