Omnifont, trainable OCR marry Today s leading OCR trend is the marriage of omnifont recognition with trainability. Even a year ago, religious wars were fought over the strengths and weaknesses of each approach. Now the battle is over; compromise has triumphed. Nearly every vendor we mention in this section showed software that contains both omnifont and trainable routines. Most of these programs can recognize most text automatically and let the user "teach" it the characters that stump it.
Caere continued an impressive streak of developments by introducing OmniPage Professional. Combining the existing OmniPage (formerly $895, now reduced to $695) with all of its formerly optional add-ons, Professional ($995) also boasts a new shape training capability that permits new and creative uses.
By working with Microsoft, Caere has managed to create a 32-bit program that runs under the 16-bit Windows 3.0 environment. This trick lets the software run more quickly. However, users pay a price for all that power: the minimum platform for OmniPage Professional is a '386 PC with 4 MB(!) of RAM.
Caere has taken trainability a step further: OmniPage Professional users can train it to recognize any symbol. After performing recognition, the program can build a table of characters, with the least well-recognized ones at the top left and with confidently recognized ones shown in alphanumeric order. The user can then give any symbol a character-string name, which will thenceforth be substituted whenever that symbol occurs.
Obvious users for this function range from recognition of mathematical formulas (the output file would contain "alpha," "beta," etc., wherever those Greek characters occurred) to very rough translations of ideographic languages (by training OmniPage Professional to substitute …