Phase and Magnitude in Normalized Images

Distortion due to complete loss or perturbation of phase or magnitude information is considered for energy normalized images. With certain statistical assumptions on the image structure, phase information is shown to be more crucial. EDICS Classification: 1.8 (Alt: 1.9, 1.7) This work was supported in part by the U.S. Air Force through DARPA grant AFOSR-89-0506 and in part by the National Science Foundation Systems Research Center. It will appear in IEEE Transactions on Image Processing in October 1994. I. Introduction The relative importance of Fourier phase and magnitude in various classes of signals has been recurring topic in signal processing literature for more than a decade [2, 4, 5, 6]. This paper considers the case of energy normalized monochrome images. Normalization of processed images to a predetermined energy before comparison is a common practice because most important visual features in an image are unaffected by moderate changes in its overall brightness. Examples ..