A method is disclosed herein for evaluating quality of an image. The
disclosed method contemplates receiving a spectral image and extracting a
plurality of pixels therefrom. The plurality of pixels are converted into
a plurality of spectral vectors, wherein each element in each spectral
vector represents a property of a respective one of N spectral bands. The
plurality of spectral vectors are then categorized into a set of M
classes. The method further includes the step of computing a mean vector
for each of the M classes based upon the spectral vectors associated
therewith. Next, spectral similarity values between pairs of the mean
vectors are computed. The distribution of these spectral similarity values
may then be analyzed in order to obtain information relevant to image
quality.