Summed squared distance error reduction by simultaneous multiprojections and applications (Q1855154)

The projections onto convex sets (POCS) algorithms are used for designing or restoring a signal, subject to a (finite) number of constraints that the signal must satisfy. A common version of the POCS algorithm is the serial cycling projection method. In the paper the POCS procedure is extended to encompass situations that could not be treated by conventional procedures. These include the possible nonconvexity of the comstraint sets, the use of projections onto sets with respect to different distance functions, and possible inconsistency of the problem. An algorithm which is a multidistance parallel projection method that generates iterates with a summed squared distance error reduction property, irrespective of the sets being convex or not, irrespective of the number of sets, and allowing multiple distance functions, as well as two application examples, is presented. The proposed method can be applied to a large variety of problems, especially complicated deconvolution-type problems, which are not easily accessible via uni-distance methods.