OCR Language Models with Custom Vocabularies

TL;DR

This paper presents an algorithm to dynamically generate domain-specific vocabularies and integrate them with general language models in OCR systems, significantly improving accuracy in specialized document recognition.

Contribution

It introduces a novel method for attaching custom vocabularies to language models at runtime and a modified beam search decoder for better hypothesis management.

Findings

Reduced word error rate in specialized domains

Effective integration of custom vocabularies with general models

Improved OCR accuracy for domain-specific documents

Abstract

Language models are useful adjuncts to optical models for producing accurate optical character recognition (OCR) results. One factor which limits the power of language models in this context is the existence of many specialized domains with language statistics very different from those implied by a general language model - think of checks, medical prescriptions, and many other specialized document classes. This paper introduces an algorithm for efficiently generating and attaching a domain specific word based language model at run time to a general language model in an OCR system. In order to best use this model the paper also introduces a modified CTC beam search decoder which effectively allows hypotheses to remain in contention based on possible future completion of vocabulary words. The result is a substantial reduction in word error rate in recognizing material from specialized domains.