Matrix-Valued Filters as Convex Programs