Image Enhancement: A Data Dependent Systems Approach

Abstract

A new image enhancement scheme based on a mathematical model obtained by data dependent systems (DDS) approach is described in this paper. A separable 2-D AR model is fitted to the image. Analysis of this model leads to the identification of modes corresponding to dominant physical features. Other intrinsic modes, inherent to the image, are highly damped and constitute difficult-to-interpret local image behavior. Although exerting a minor influence on the image intensities, they hinder a clear perception of the image. In order to enhance the image, these modes must be filtered out. ARMA image enhancement filters are formed using the major inherent modes. Residuals, the part of the image not modeled by this ARMA filter, comprise the enhanced image. This approach can also be used to selectively enhance the desired image features. Examples illustrating clear enhancement of a real image, with natural degradations created by shadows and other artifacts as well as artificially added noise, are given.