Getting Started | Documentation | Glish | Learn More | Programming | Contact Us |
Version 1.9 Build 1367 |
|
Apart from the dirty image, the simplest class of Image Solvers are those that use the principle of Projection Onto Convex Sets. POCS is a simple but very general iterative algorithm. If one wants to solve a linear equation AX = Y then one uses an iterative algorithm given by successive and repeated applications of various projection operators:
where the projection operators Ti impose some form of a priori constraint e.g. positivity, finite support. CLEAN is a POCS algorithm. A POCS algorithm is guaranteed to converge for convex set.
In the context of the GI measurement equation, Y - AX is the residual image . The projection operators are open to choice. There are a number of possible operators:
I 0 | (5) |
I2 - Q2 - U2 - V2 0 | (6) |
This describes a hypercone in I, Q, U, V space. For points outside the cone, projection onto the surface of the cone can be done in a number of ways. The simplest operator, Tcone, sets to zero all elements outside the cone. Alternatively one could boost the I component until the point lies inside the cone, Tboost. The most appropriate operator, Tshrink, is probably that which projects perpendicular to the I axis so as to shrink Q, U, V to be less than or equal to I.
A terser, and perhaps useful equation expressing this constraint is that:
where the matrix M is given by:
One can also think of polarized radiation in terms of the coherence matrix:
in which case the equivalent statement is that this matrix be positive semi-definite, and so the product of eigenvalues is non-negative:
0 | (10) |