Application of linear prediction characteristics to planar shape classification

Several different parametric representations of planar shapes for classification purposes are presented. The new approaches are based upon the linear prediction technique and are alternatives to the well-known circular auto regressive (CAR) model approach. The proposed representations include the impulse response function, autocorrelation function, and the cepstrum function. They can be used to represent both closed and open planar curves and are invariant to rotation, translation, and scaling. The effectiveness of the new representation for classification of closed planar curves is examined. Their performance is compared and the results are reported