MATEs: Multilingual Automatic Translation Engine

Phase II of the MATES project will extend our current system for road sign detection, recognition and translation on handheld devices in several novel and innovative directions. Our core application will be document triage, where the soldier obtains a gist translation or keyword profile of a questioned document using a camera-enabled PDA. We will apply the technology to Arabic language documents by integrating new recognition modules; but continue to support the language components developed in Phase I. We build on modules developed in Phase I, with proposed extensions for: * document capture and mosaicing, * document enhancement including blur reduction, * dual-mode word spotting and full translation using robust shape coding for Arabic, * novel dictionary nearest neighbor search useful for dealing with spelling errors and infix Arabic morphology, and * enhanced feedback to the solider as part of the user interface. We will perform user needs assessments for new features and periodic performance and user evaluations to mark progress. This dual use technology will address many of the needs of the government and the commercial mobile camera phone industry. The modular component architecture that we have developed, along with cross-platform coding allows us to easily port the system to camera-based mobile phones.