MODELING THE SENSITIVITY OF MOMENT INVARIANTS IN A RECOGNITION SYSTEM

We derive a sensitivity analysis for moment invariants of multidimensi onal distributions. These invariants have many uses in computational s ystems and have recently been used for illumination-invariant recognit ion in color images. In this contest, the sensitivity analysis predict s the response of moment invariants to partial occlusion. Using the re sults of the sensitivity analysis, we develop a novel surface represen tation called the invariant profile, which captures color distribution and spatial information while remaining invariant to the spectral con tent of the scene illumination. Unlike for previous representations, t he recognition of invariant profiles does not require illumination cor rection. Ne demonstrate the sensitivity analysis and the use of invari ant profiles for recognition with a set of experiments on color images . (C) 1998 Optical Society of America.