3D Reconstruction from Two Orthogonal Views Using Simulated Annealing Approach

The technique of 3-dimensional (3D) reconstruction from two orthogonal images is mainly used in ventricle or vessel reconstruction. In the literature, the 3D object is considered to be a stacked 2-dimensional(2D) slice set and 3D reconstruction can be achieved by constructing a stack of 2D slice reconstructions, by rendering each slice into two 1-dimensional profiles corresponding to a pair of rows obtained from the segmented projections. Previous work has modelled each slice as a 2D Markov-Gibbs random field, and a simulated annealing approach is used to solve the 2D reconstruction problem based on two infinite-source orthogonal views. The current work applies this approach to 3D space in particular. We find this approach can be applied to more wide application, rather than only in cardiac reconstruction. The results obtained by reconstructing binary and multiple objects of the human voxel phantom from two orthogonal projections, yield low reconstruction errors and the shapes of the reconstructed objects are observed to match the original objects to a high degree.