Invariant distance measures for planar shapes based on complex autoregressive model

摘要

Several distance measures (Euclidean distances between complex autoregressive coefficients or complex partial correlation coefficients, log-likelihood distance, and complex power cepstrum distance) between planar shapes are presented on the basis of a complex autoregressive model. The measures are invariant to translation, rotation, and scale of patterns. Since the concept of “shape” is essentially by independent of such transformations of patterns, these measures are suitable for classification, identification, or clustering of planar shapes, like unsupervised applications. The properties of the proposed distance measures are shown with their experimental comparisons.