Abstract
In the first half of this paper, we examined psychological validity of two measures derived from singular value decomposition of‘0-1’matrix in which a Kata-Kana letter was quantitized. Variance of extracted eigenvalues reasonably predicted recognition accuracy of a letter, and variance of contributions of principal components correlated with complexity and regularity ratings of a letter. In the last half, we constructed the pattern recognition model implementing pre-process by singular value decomposition and also feature-sampling-process by Walsh-Hadamard transformation. Similarity structure output through the model was quite similar to the one obtained from the psychological experiment.
