PUB - Publikationen an der Universität Bielefeld

With the increasing computational support for collaborative work-environments electronically enhanced whiteboards have been developed to serve as automatic meeting assistants. The most flexible of these systems use cameras to observe the whiteboard, and, therefore, do not require the use of special pens or erasers. However, currently these systems are only capable to interpret some special graphical symbols and can not produce transcripts of the documents written on them. As a major advancement beyond the state-of-the-art we propose a system for automatic video-based reading of unconstrained handwritten text from a whiteboard. Text lines are extracted from the captured image sequence using an incremental processing strategy. The recognition results are then obtained from the text-line images by off-line techniques and a segmentation-free statistical recognizer. We will present results on a writer independent unconstrained handwriting recognition task showing that handwriting recognition can successfully be applied to automatically reading texts from whiteboards.