A distortion invariant feature extraction algorithm used with associative memory classifier

The authors propose a feature extraction algorithm which can extract feature vectors not only invariant to geometric distortions but also suitable for the associative memory classifier developed in the authors' previous work (1991). In addition to the deformation invariant property, the suitability of feature vectors for a neural network system is of a major concern in the feature extraction process when the latter is utilized as a pattern classifier. Under this circumstance, data reduction becomes less important since neural networks are intrinsically appropriate for accommodating large amounts of data. These opinions were justified by experiments on the recognition of a set of multifont Chinese characters with the associative memory classifier using the features extracted by the algorithm.<>