Date of Award
Doctor of Philosophy
Pattern recognition approach is used in this research as an alternative to univariate and rule-based systems for recovering source text encrypted in monoalphabetic and polyalphabetic substitution ciphers. The approach uses a multivariate statistical technique called discriminant analysis to classify encrypted characters as source text symbols. It requires the generation of invariant measurement vectors for all groups of characters from the source and the encrypted text to classify and to identify characters. The measurement vectors consist of the values of variables generated from one-graph and digraph structures of the text.;The research suggests that quadratic discriminant analysis, which requires more time and space in the process of computation is not better than linear discriminant analysis. Furthermore, the stepwise linear discriminant procedure is useful in selecting variables in the measurement vector and in reducing the number of redundant variables. The use of digraph structures of text improves the total and individual percentages of correctness for the groups of characters in comparison to the use only of one-graph structures. They also improve the results of recovering the plain or source text.;The research suggests that the use of trigraph structures will further improve the percentages of correctness and should be taken into consideration for continuing this research.
Syamsun, Muhammad, "Pattern Recognition In The Structure Of Strings Of Characters Using Multivariate Statistical Analysis" (1994). Digitized Theses. 2436.