Abstract: This paper is dedicated to the problem of two-dimensional (2D) shape recognition from the point of view of possible optimization regarding feature extraction and classification methods. Moment invariants and stochastic AR models are considered as feature extraction methods. For classification, we analyze performance of the Bayesian parametric and nonparametric classifiers, as well as of the multilayer perceptron, applied as a nonparametric classifier. Experimental analysis is based on real data. Evaluation of the considered methods is done on the basis of the Bayes error estimates, calculated on the corresponding data sets.
Key words: Patern recognition, AR models, moments.