Polynomial transforms as a tool for adaptive image processing
Jbos Jean-Bernard Martens · TU/e Research Portal · 1992
Previous work at the Institute for Perception Research has resulted in a new model for representing images. called a polynomial transform. This transform mimics properties of the early stages of (human) vision. It detects luminance changes in images and represents them by a sum of basic patterns. i.e., localized polynomials. The localized nature of the basic patterns in a polynomial transform makes the latter ideally suited for adaptive image processing. This paper presents a general view of adaptive image processing by means of polynomial transforms. It is shown that the two basic tools in adaptive processing are transformation rules and control variables, both of which can be easily implemented using polynomial transforms. The transformation rules describe how important image degradations such as sampling, blurring and noise affect the coefficients in a polynomial transform. By inverting these transformations, we can obtain interpolation, deblurring and noise reduction. The control variables are estimated properties of the underlying (original) image and are typically used to vary the processing for different positions in the image. Some specific applications of this general approach, e.g., in adaptive noise reduction and image deblurring. have been published earlier.