“Programmic filters”: A framework for the design of high-performance nonlinear filters for low-level picture processing
Oichi Atoda, Koichi Okano, Masaki Tomisawa, Hitoshi Tamura · Systems and Computers in Japan · 1997
Into the chaotic space of nonlinear spatial filters for low-level picture processing such as smoothing or textural feature extraction, we introduce a design framework in which a designer converts heuristics into filter functions, similarly to the activity of a computer programmer formulating program concepts in the framework of a programming language. First, to assure uniform gain and bias change, filters are constrained to respond in the same manner as their linear counterparts. Then it is shown that a “quasi-linear” filter is generated from successive application of several rules similar to a sentence from a parse tree. Filters so generated are called programmic filters. Some of the rules allow Boolean functions, which are suitable to heuristic thought. Thus, the set of those rules constitutes the filter design framework. Examples of high-performance programmic filters including an edge-preserving texture-reject filter are presented. © 1997 Scripta Technica, Inc. Syst Comp Jpn, 28(3): 71–81, 1997