Universal Analytical Forms for Modeling Image Probabilities

Seeking universal probability models for image representations, we employ a spectral approach where the images are decomposed using several bandpass lters, and probability models on the lter outputs (or spectral components) are imposed. We apply a (two-parameter) analytical form, introduced in [9] and called a Bessel K form, for modeling the marginal probabilities of these spectral components, and demonstrate their t to the observed histograms for video, IR, and range images. A relationship between the Bessel parameters and certain characteristics of the imaged objects is established. Using an expression for L-metric on the set of Bessel K forms, we suggest a pseudo-metric on the image space for quantifying image similarities/di erences. Some applications, including clutter classi cation and pruning of hypotheses for target recognition, are presented.