Learning image invariants with the Volterra connectionist model

The theory of moment invariants is used to look at the type of geometrical image invariances, such as translation and rotation, that can be learned by the Volterra connectionist model (VCM). The VCM uses linear adaptive filter theory on a polynomial extension of the input vector; it is shown that it can learn to classify patterns independent of translations and rotations. Invariants to other transformations, such as changes in contrast, scale, and affine transformations, can be represented by the ratio of two weighted extended vectors, and can be learned using principal component analysis