An introduction to robust shape classification

Robust shape classifiers are compared, and it is found that conventional techniques based on the sample auto-covariance function suffer catastrophic reductions in performance in outlier contaminated data. However, robust procedures suffer much less degradation, with the robust spectral approach giving the best performance. The use of lag selection in the classification phase may be of independent interest and is related to the use of the smoothed periodogram in time series discrimination. This approach is well suited to problems where sensitivity to clutter is important. Typical examples are fault identification, or the recognition of new objects entering a domain.