An operations research approach to the modeling and analysis of different feature sets proposed for human perception of capital letters

Abstract A Markov chain feature transition model of capital letter recognition is presented for synthesizing the empirically derived 26 × 26 capital letter confusion matrices. The transition probabilities associated with the features in the model are used as solution vector in an optimization problem, the objective being to minimize the sum of per cell squared differences between the empirical confusion matrix and the synthesized matrix. Three proposed feature sets are used on conjunction with the model to synthesize three empirical confusion matrices. Variants of the model are considered and results are tabulated and analyzed.