Linear estimation theory is developed in the context of object reconstruction from data obtained by a general shiftvariant imaging system. The formalism adopts nonstationary first- and second-order statistics of the object and noise classes as a priori information. In addition, a metric for system optimization that depends on the a priori information is presented. The role of this a priori information as derived from several different training sets is then studied with respect to reconstruction performance for various noise levels in the data, using a tomographic codedaperture system as the model. In a separate experiment, a simple coded-aperture system is optimized to a particular object class, and the results are compared with those from an earlier optimization experiment.
|Number of pages
|Journal of the Optical Society of America A: Optics and Image Science, and Vision
|Published - Mar 1988
ASJC Scopus subject areas
- Electronic, Optical and Magnetic Materials
- Atomic and Molecular Physics, and Optics
- Computer Vision and Pattern Recognition