Generalized affine moment invariants for object recogn

Abstract

This paper introduces a new way of extracting affine invariant features from image functions. The presented approach is based on combining affine moment invariants (AMI) with multiscale invariants, in particular multiscale auto convolution (MSA) and spatial multiscale affine invariants (SMA). Our approach includes all of these invariants as special cases, but also makes it possible to construct new ones. According to the performed experiments the introduced features provide discriminating information for affine invariant object classification, clearly outperforming standard AMI, MSA, and SMA