Abstract

Image moments are invariant under certain coordinate transformations. In practical use, the classical invariant moments are sensitive to noise and to measurement errors if the moments are generated by optical methods. The effects of noise on image moments decrease with the moment order. Low-order invariant moments may be expressed in terms of the radial moments of circular harmonic functions or extended complex moments. For the purpose of invariant pattern recognition, these moments may be used for image description followed by a statistical classification in feature space. They may also be used for image normalization followed by a classification using image correlation. For image normalization the use of low-order moments of circular harmonic functions yields an image mean direction which is more robust against background noise than the classical principal axes.

© 1986 Optical Society of America