抄録
This paper represents the possibility of pattern recognition by multi-level multi-valiate analytic method. We argue about the equality of information condense and the meaning of eigenvalues in amultivaliate analysis. Then validity of the method was exsamined in way of pattern recognition. 508 of 20 dot pattern(_??_, _??_, _??_, A, C, T)in 16×16 mesh were valuated meaning primitives as gravity, centrality, vertical and holizental view point. Next, each were condensed by factor analysis and associated by principal component analysis. The results showed high discrimination. It suggests that multi-level multi-valiate analytic method is an effective way of information condense on recognition.
