This report presents 2D radiographic reconstructions using the method of Prior Singular Value Decomposition (pSVD). The technique utilizes (typically minimal) prior knowledge of the object to enhance reconstructions by the inclusion of projection operator null space vectors. Examples of prior knowledge illustrated in this report are that the object density is: (1) non-negative; (2) bounded above and below; (3) a known constant; and (4) of a known set of discrete values. Several test objects and simulated noisy data are used to illustrate the method. One set of actual proton radiographs taken at LANSCE is used to verify the method. The radiographed object (BCO4) was viewed from 31 equally spaced angles and remained static. The pSVD reconstructions are shown to be superior to the standard filtered back projection reconstructions. It is found that known constraints improve reconstructions by any reasonable metric. Even for the simple constraints discussed here, the null space enhancements can be significant. In some cases sparse noisy data is sufficient for essentially exact reconstructions. The implications for experiments with limited data is obvious. Natural extensions to 3D reconstructions and dynamic constraints are discussed. Appendices cataloge many reconstruction examples of both standard test objects and the BCO4.

LA-UR-03-5937