Indirect Symbolic Correlation Apporoach to Unsegmented Text Recognition

During the last twenty years, most recognition engines for difficult to segment scripts have been built around Hidden Markov Models (HMMs). Parametric recognizers for unsegmented signals, like HMMs, are hard to train. In contrast, non-parametric classifiers, like Nearest-Neighbor, require only a labeled reference list. In this paper, we provide preliminary results in support of an entirely new method for non-parametric classification of unsegmented text. Indirect symbolic correlation is a general method for bringing lexical context into the recognition of unsegmented signals that represent words or phrases in printed form. It is applicable wherever segments of lexically labeled reference signals can be compared to unlabeled signals. The signal need only preserve the ordering of the alphabetic units within a word (and of words within a phrase).