Video cameras are becoming inexpensive, and also very small, which makes
it reasonable to start integrating them into laptop computers as a
standard feature. Aside from taking snapshots and enabling video conferencing,
we believe there are a number of other uses for such a camera. One application
we have focused on is automatically reading business cards presented to the
system.
In our prototype implementation the user holds a document (in this case a business card) in front of the camera. When the image is large enough and roughly aligned with a set of provided fiducials, the system snaps a number of gray scale images. These are then combined using a hyper-resolution scheme (to deblur the text) and then further enhanced in terms of overall contrast and edge definition. Finally, the resulting image is passed to a commercial OCR package designed specifically for business cards, which parses the text into the appropriate fields of a database. The system currently handles about 35% of the cards flawlessly, and up to 85% with only minor errors. We are investigating additional ways to improve this performance by exploiting known pixel geometries and pre-correcting for text slant and stroke width variation. |
| Contact: Rick Kjeldsen | Last updated: 6/12/02 | ||
 |
 |
 |
 |