Computational imaging systems: joint design and end-to-end optimality.

A framework is proposed for optimal joint design of the optical and reconstruction filters in a computational imaging system. First, a technique for the design of a physically unconstrained system is proposed whose performance serves as a universal bound on any realistic computational imaging system. Increasing levels of constraints are then imposed to emulate a physically realizable optical filter. The proposed design employs a generalized Benders' decomposition method to yield multiple globally optimal solutions to the nonconvex optimization problem. Structured, closed-form solutions for the design of observation and reconstruction filters, in terms of the system input and noise autocorrelation matrices, are presented. Numerical comparison with a state-of-the-art optical system shows the advantage of joint optimization and concurrent design.