Statistical signatures of scale and amplitude variations in fourier spectra

Fourier spectra are often used as input data when classifying images by statistical transforms. Optical biases such as variations of the total energy level in the images, or undetected changes of scale from one image to the other interfere with the statistical features of interest. Such biases have statistical signatures which are investigated in this study. Characteristic patterns observed by principal component analysis are shown, as well as a possibility of a quantitative relationship between the magnitude of the biases and the distances along the patterns.