Automated Document Pre-Processing Technologies for Optical Character Recognition (OCR)

Todays Army routinely supports joint interagency, intergovernmental, multinational missions where critical information is often written in foreign languages. To address this issue, the U.S. Army has developed the Machine Foreign Language Translation System (MFLTS) that aims to provide accurate and rapid translation of foreign languages to English. Nonetheless, the performance of the Optical Character Recognition (OCR) in MFLTS is often limited by degraded or noisy documents encountered in the operational environment. Labor-intensive manual pre-processing of the document images is often required in the current operational environment, which motivates us to develop automated document image pre-processing software to improve the OCR process without depending on the skill level of human operators and to increase the overall efficiency of the MFLTS task by reducing human operators workload.Leveraging on our prior development in handwritten document recognition and computer-vision-based airport surveillance system, machine-learning-based pre-processing technologies will be developed to address the degradation in OCR accuracy in operational environments. A feasibility demonstration of the proposed document pre-processing software will be provided by the end of the Phase I.