HIERARCHICAL BAYESIAN APPROACH TO RECONSTRUCTION FROM PROJECTIONS OF A MULTIPLE OBJECT 3-D SCENE.

A suboptimal algorithm for the reconstruction of multiple 3-D objects from a limited and noisy set of projections is presented. The knowledge about the properties of the objects and the imaging process is embedded in a mathematical probabilistic model, converting the problem into an estimation problem, posed directly in projection space. The proposed hierarchical algorithm, alleviates the computational difficulties, by a massive data reduction in each of its steps. The algorithm may be also applied to general image segmentation or boundary detection problems with multiple, possibly overlapping objects.