Business cards - forthcoming optical character recognition software such as Cognitive Technology's Business Card Reader and Corex's CardScan facilitate storage of business card data - one of several articles on Personal Data Interchange technologies

RELease 1.0, Sept 23, 1993 by Jerry Michalski

Even if many vendors agree on standards and embed compatible inflated transceivers in their next-generation gadgets, it will still take five years for them to be commonplace, and maybe ten for them to be the norm (remember the day you realized you couldn't do business any more without a fax?). Meanwhile, everyone carries paper business cards, and people spend countless hours typing in the information they contain.

Although this problem has been around for a ions time and a few products that can scan and read business cards have been on the market, principally in Asia, where pictographic writing has made scanners and imaging popular, their price/performance tradeoff has not attracted many buyers. Now, several startups are hittint the US market with business-card reading systems street priced around $300. The burst of activity in business-card OCR is partly due to the recent availability of inexpensive greyscale scanners. Black-and-white scanners often don't provide enough resolution for accurat, recognition; faxes, typically at 200 dpi resolution, are also poor fare for OCR engines (which is a shame, since you could fax your new business cards in from anywhere).

Two companies, Cognitive Technology and Corex, are developins products that are similar in many respects. Both use scanners, which they drive automatically as the user feeds in business cards, one by one (separating multiple cards in a single scanned image is a wish-list feature). Both adjust contrast and compensate for skewed images, and store the full card image for retrieval later. Then the software identifies regions on the business card, decides which ones to recognize, invokes an OCR (optical character recognition) engine, infers a mapping of recognized fields to expected fields and presents the results to the user. Both products run under Windows; neither is really a RIM, though Corex tries to be one, with its Rolodex-like user interface.

Will CuneiForm read clay tablets?

Cognitive Technologies is likely to field its card reader first. The company was rounded in 1990 by Russian Yefim Schukin and Denis Coleman, who co-rounded C&E Software with Gordon Eubanks, which later acquired Symantec. Schukin. a theoretical physics Ph.D., emigrated to the US in 1973 and worked at the Stanford University AI lab for two years. After teaching and consulting, he formed a company around a team of Russian scientists in 1988 to develop AI applications. Along with some US scientists, they form the core of Cognitive Technologies. They have developed image-processing algorithms and a character-recognition product called CurteForm, which is the basis for Cognitive's Business Card Reader (BCR).

Cognitive is actively licensing its OCR engine. For example, Corel licensed an early version in 1992 for integration with CorelDraw; EDS has licensed it to recognize the preprinted text on preapproved GM credit-card applications (they think it will be faster than coding and sorting them). Dextra and several other vendors of small, inexpensive vertical-format scanners. will bundle Cognitive's software with their machines. And most visibly, Lotus has Joined Cognitive in an agreement to link the product to Organizer, Lotus' popular RIM. The agreement should give Cognitive much more exposure and Organizer a valuable edge.

CuneiForm optionally uses Brayscale scanning for better accuracy. It also uses adaptive thresholding, which helps compensate for colored backgrounds or faint writing and other business card irregularities. After the scannins and recognition process described above, the system displays the card image and recognized text on the same screen. Users can correct misrecognized or misplaced text by drawing rectangles around its image and dragging it to the right field; then the recognition engine tries again. (Where the spelling was fine but the field was wrong, our instinct was to grab the already-recognized text and move it to the correct field, but that's not the way the software works.)

Long run. Schukin envisions his engine recognizing any printed matter 'from travel receipts to checks. Where once you might have photocopied the checks for your files, you can now recognize their contents and store the image at the same time.

Corex: spelunkers of the modern age

Boston-based Corex is starting with a similar offering, but has different long-term plans. It expects to ship an automatic business-card reading system called CardScan, which puts recognized data in a Rolodex-style format on a PC screen, by the end of 1993. In 1994, Corex will also offer another program called People & Places, which will help sift address information from documents such as business letters and mailing lists.

Corex licenses other vendors' OCR engines. Instead of focusing on image-processing algorithms, it specializes in downstream tasks such as parsing and normalizing information, as well as synchronizing address books with other sources of address information.

Where Cognitive is looking to expand into other paper sources of information such as receipts, Corex is looking to leverage its parsing and file-synchronization capabilities with sources already inside the computer network -- data mining/refining the address information. This should make it appealing to mobile computer users. It is also working on strong interoperability with other pc applications. For example, a user writing a letter in a word processor will be able to type the recipient's name, hit a People & Places icon, and insert the full mailing address automatically.