Development of a Hand-Held Camera for Inkprint to Text Translation

A handheld camera which translates inkprint into computer readable text has a wide variety of potential applications, such as the input device for a voice output reading machine for the visually impaired. Commercial optical character recognition systems typically operate on controlled input and read with high speed and accuracy. As an input device for a voice output reading machine, much slower speeds of 200 words per minute are acceptable, but the text input is much less controlled in terms of quality, type size, style and format. In order to acquire text from a complex format which may include multiple columns, pictures, and graphs, operator control of the scanning is important. Automatic control of threshold and magnification combined with user control of the scanning sequence based on direct feedback from the camera image offers a potentially efficient input structure. Automatic thresholding and magnification control algorithms which work well on newsprint and good quality type are presented. Based on these results, spatial resolution and quantization requirements can be established for a system which will read text with an order of magnitude variation in size. Direct conversion auditory and tactile feedback for user control of scanning are considered.