Handwriting recognition

While the keyboard is the traditional means for entering text into a computer system, both designers and users have long acknowledged the potential benefits of a system where people could enter text using ordinary script or printed handwriting and have it converted to standard computer character codes (see characters and strings). With such a system people would not need to master a typewriter-style keyboard. Further, users could write commands or take notes on handheld or “palm” computers the size of a small note pad that are too small to have a keyboard (see por-table computers). Indeed, such facilities are available to a limited extent today.

A handwriting recognition system begins by building a representation of the user’s writing. With a pen or stylus system, this representation is not simply a graphical image but includes the recorded “strokes” or discrete movements that make up the letters. The software must then create a representation of features of the handwriting that can be used to match it to the appropriate character templates. Handwriting recognition is actually an application of the larger problem of identifying the significance of features in a pattern.

One approach (often used on systems that work from previously written documents rather than stylus strokes) is to identify patterns of pixels that have a high statistical correlation to the presence of a particular letter in the rect-angular “frame” under consideration. Another approach is to try to identify groups of strokes or segments that can be associated with particular letters. In evaluating such tenta-tive recognitions, programs can also incorporate a network of “recognizers” that receive feedback on the basis of their accuracy (see neural network). Finally, where the identity of a letter remains ambiguous, lexical analysis can be used to determine the most probable letter in a given context, using a dictionary or a table of letter group frequencies.

Implementation and Applications

A number of handheld computers beginning with Apple’s Newton in the mid-1990s and the now popular Palm devices and BlackBerry have some ability to recognize handwrit-ing. However, current systems can be frustrating to use because accuracy often requires that users write very care-fully and consistently or (as in the case of the Palm) even replace their usual letter strokes with simplified alternatives that the computer can more easily recognize. If the user is allowed to use normal strokes, the system must be gradu-ally “trained” by the user giving writing samples and con-firming the system’s guess about the letters. As the software becomes more adaptable and processing power increases (allowing more sophisticated algorithms or larger neural networks to be practical) users will be able to write more naturally and systems will gain more consumer acceptance. (One step in this direction is the Tablet PC, a notepad-sized computer with a digitizer tablet and a stylus and handwrit-ing recognitions software, included in Windows XP and expanded in Windows Vista. Programs such as Microsoft OneNote use handwriting recognition to allow users to incorporate handwritten text into notes that can be orga-nized and quickly retrieved.)

Currently, handwriting recognition is used mainly in niche applications, such as collecting signatures for deliv-ery services or filling out “electronic forms” in applications where the user must be mobile and relatively hands-free (such as law enforcement).