Discrete quasi-eigenfunction approximation for AM-FM image analysis

Joseph P. Havlicek, David S. Harding, Alan C. Bovik

Abstract: We introduce a multidimensional quasi-eigenfunction approximation useful for characterizing and analyzing the responses of discrete linear systems to inputs modeled as AM-FM image functions. A new theorem is presented which provides a global bound on the approximation error. We demonstrate how the approximation may be used to develop discrete AM-FM demodulation algorithms, and show dramatic demodulation examples where the essential structure of natural images is captured using only a small number of AM-FM components.