A normalized, power-weighted averaging filter (NPF) is a very good approximation to the well-know local maximum and minimum filters along the object edges and offers noise reduction in foreground and background regions. This favorable combination turns out to be very effective in image smoothing, edges detection and image sharpening. It offers a clear improvement over the existing operators based on true max-min filtering. Our filter can be implemented very efficiently by a table lookup (for the power) and two (separable) convolutions.
Lucas J. van Vliet