Convergence properties of multi-dimensional stack filters
Peter Wendt
Electronic Imaging: Advanced Devices and Systems 1990
We explicitly obtain, for K(x, y) totally positive, a best choice of functions u1, ..., un and v1, ..., vn for the problem minui, vi (∝01 (∝01 |K(x, y) - ∑i = 1, n ui(x) vi(y)| dyp dx) 1 p, where ui ε{lunate} Lp[0, 1], vi ε{lunate} L1[0, 1], i = 1, ..., n, and p ε{lunate} [1, ∞]. We show that an optimal choice is determined by certain sections K(x, ξ1), ..., K(x, ξn), and K(τ1, y), ..., K(τn, y) of the kernel K. We also determine the n-widths, both in the sense of Kolmogorov and of Gel'fand, and identify optimal subspaces, for the set Kr,v = {f(x) = ∑ i=1 raiki(x) + ∫ 0 1K(x,y)h(y)dy, (a1, ..., ar)ε{lunate}Rr, {norm of matrix}h{norm of matrix}p≤1}, as a subset of Lq[0, 1], with either p = ∞ and q ε{lunate} [1, ∞], or p ε{lunate} [1, ∞] and q = 1, where {k1(x), ..., kr(x), K(x, y)} satisfy certain restrictions. A particular example is the ball Br,v = {f} in the Sobolev space. © 1978.
Peter Wendt
Electronic Imaging: Advanced Devices and Systems 1990
Richard M. Karp, Raymond E. Miller
Journal of Computer and System Sciences
Timothy J. Wiltshire, Joseph P. Kirk, et al.
SPIE Advanced Lithography 1998
Michael E. Henderson
International Journal of Bifurcation and Chaos in Applied Sciences and Engineering