Shape recognition using spectral features

The classiication of planar shapes using spectral features is presented in this paper. The contour of a planar shape is represented by magnitude and phase of radial vector drawn from a centroid, and they are modeled by an autoregressive process. The spectral features are extracted from the least squares estimators of model parameters , and planar shapes are classiied by an artiicial neural network. In the experiment with contours of aircrafts, machine parts and handwritten numerals, more than 96 percent of shapes are correctly classiied.