Document image decoding approach to character template estimation

Template Estimation 1 Gary E. Kopec2 Xerox Palo Alto Research Center Mauricio Lomelin3 Microsoft Corp. November 29, 1995 Abstract This paper develops an approach to supervised training of character templates from page images and unaligned transcriptions. The template estimation problem is formulated as one of constrained maximum likelihood parameter estimation within the document image decoding framework. This leads to a two-phase iterative training algorithm consisting of transcription alignment and aligned template estimation (ATE) steps. The maximum likelihood ATE problem is shown to be NP-complete and thus a number of simple suboptimal solutions are developed. The training procedure is illustrated by its use in creating a document-specific decoder for high-accuracy transcription of a large (400 page) text document. Depending on the language model used, the decoder character error rate is a factor of 7–20 less than that of a commercial omni-font OCR program; the best case error rate is 0.036%.