Recognition of hand-drawn geometrical figure using multiple discriminant analysis
M. Yabuuchi · The Japanese Journal of Ergonomics · 1972
The purpose of this paper is to examine an algorithm of the statististical pattern recognition processing, which can be considered as consisting of three processes: normalization, selection of feature measurements, and classification. The algorithm was applied to hand-drawn simple geometrical figure recognition. The pattern samples used for two experiments were obtained by asking subjects to draw figures in a 10cm-square (L-cond.) or in a 5cm-square (S-cond.). As to both conditions, a simple normalization technique was deviced and estimated by testing equality of mean vectors and covariance matrices. An idea of multivariate statistical control was utilized to detect abnormal input patterns. The good recognition accuracy was obtained in the results of two experiments.