Status and a technology roadmap for document recognition

Intelligence agencies collect vast quantities of data from diverse sources including video, still images, audio recordings, and documents. To extract useful content to "connect the dots," they need to employ computer vision, speech recognition, and document recognition technology. This talk takes stock of the document image analysis portion of the challenge. In this talk, we: (1) survey the state of the art, exploring what works, and what doesn't; (2) consider the drivers and leading edge of both the commercial document recognition industry and current academic research; (3) touch on major trends and a dozen hot topics; (4) consider how and where progress gets made, and how to move it from the laboratory to applications; and (5) discuss major fundamental challenges to the field, and recommend process steps that would maximally leverage R&D investment.