Applications of computer vision to document capture and processing from cameras

Maurizio Pilu (HP Labs Bristol, UK)

Capturing an image of a document from a hand-held camera presents interesting and unexpected problems such as illumination gradients and specularities , cast hand shadows, motion blur, perspective distortion, page curl, poor image quality, stiching of document portions, and HCI issues. In this talk I will present some of the novel techniques we developed to specifically address these issues. During the talk I will also give a live illustration of some of the techniques discussed by showing a text capture and translation demo running on a hand-held PDA and a demonstration of an extremely robust document stitching method.