"Machine-based handwriting recognition has been studied now for more than a century. In 1910, Hyman Goldberg proposed recognizing handwriting using electically conducting ink. Since then, the subject of handwriting recognition has grown and flourished. Handwriting recognition is essential to major economic activities, such as cheque processing and mail sorting, and is a standard feature on many mobile electronic devices.
There is by now a vast literature on the subject of handwriting recognition by computer, divided between 'off-line' and 'on-line' recognition. Off-line recognition takes a static image of some handwriting and produces text. The input is typically an image which may involve background noise, digitization artifacts and distortion. On-line recognition takes motions and other events, such as button presses, pen up and pen down, and produces text. A variety of capture devices may be used, including digitizing tablets, screen overlays or cameras. The captured pen movements and related events may be called 'digital ink' regardless of the source, and which may be stored and transmitted in a number of ways, including InkML. On-line recognition is often regarded as an easier problem because the writing order is given, the identification of the input is evident and mis-recognitions can be corrected. On the other hand, processing time becomes a constraint and there is no forward context."
- from Stephen M. Watt. Polynomial Approximation in Handwriting Recognition, pp. 3-7, Proc. 4th International Workshop on Symbolic-Numeric Computation, (SNC 2011), June 7-9 2011, San Jose, California, ACM Press.
Image from Redmond Pie.