Handwriting recognition

from Wikipedia, the free encyclopedia

Handwriting recognition is a form of pattern recognition in which handwritten characters or words are to be recognized. A distinction is made between on-line and off-line handwriting recognition. In on-line handwriting recognition, the information on how a word was written, i.e. in which order the lines were drawn, can be used for recognition. In the case of off-line handwriting recognition, on the other hand, only the end product is available as an image. Therefore, off-line handwriting recognition is a branch of text recognition , but on-line handwriting recognition is not.

General functionality

A handwriting recognizer can use several steps to recognize handwriting:

  • Preprocessing: The removal of superfluous (too high a resolution) or even incorrect information (image errors)
  • Segmentation : The breaking down of the input into smaller parts, such as B. breaking down words into letters or numbers into digits
  • Feature extraction or generation: Less, more important information is generated from the many small pieces of information. In the on-line handwriting recognition system, the information on where the pen was at what point in time can be generated from the large amount of information about the curvature of the writing curve at different points.

In a further step, the characters are then recognized using machine learning techniques .

On-line handwriting recognition

On-line handwriting recognition requires a touch screen or other digital device to record the writing while writing. A wide variety of input devices such as smartphones and tablets with fingers or input pens or graphics tablets can be used. These input devices provide information about where the input pen is located. Some devices provide this information even if the pen is not on the touchscreen, but just above it. Depending on the technology used, information on how much pressure was exerted is also supplied in addition to the coordinates of the input.

On-line handwriting recognition is currently much less common than off-line handwriting recognition.

On-line handwriting recognition has been studied scientifically since the 1950s. (In 1958, the “Stylator” was introduced as the first device with which it was possible to record on-line information relating to handwriting recognition.)

Pre-processing in online handwriting recognition can involve a number of different steps:

  • Finding and eliminating "wild points", ie points that arise from hardware errors.
  • The interpolation of the measured points and the associated recalculation of points.
  • Scaling and moving the input.
  • Finding a baseline and, if necessary, correcting the writing so that it lies on a straight baseline.
  • Finding and, if necessary, correcting the inclination of the letters.

Off-line handwriting recognition

Off-line handwriting recognition has numerous areas of application, such as handwriting recognition on forms, transfer receipts and addresses on letters.


Handwriting recognition has an active community studying it. There are various conferences such as the ICFHR or the ICDAR which serve this purpose.

See also

Individual evidence

  1. Brakensiek, A. and Kosmala, A. and Willett, D. and Rigoll, G .: Comparison of different statistical modeling methods for on- and off-line handwriting recognition . Springer Berlin Heidelberg, 1999, ISBN 978-3-540-66381-2 , p. 70-77 .
  2. Joachim Schenk, Gerhard Rigoll: Man-machine communication . Springer Berlin Heidelberg, 2010, ISBN 978-3-642-05456-3 , pp. 151-164 .
  3. ^ Tappert, CC and Suen, CY and Wakahara, T .: The State of the Art in Online Handwriting Recognition . In: IEEE Trans. Pattern Anal. Do. Intelligent tape 12 , no. 8 , 1990, pp. 787-808 , doi : 10.1109 / 34.57669 .
  4. ^ TL Dimond: Devices for Reading Handwritten Characters . In: Papers and Discussions Presented at the December 9-13, 1957, Eastern Joint Computer Conference: Computers with Deadlines to Meet . 1958, p. 232--237 , doi : 10.1145 / 1457720.1457765 .
  5. Martin Thoma: On-line Recognition of Handwritten Mathematical Symbols . November 29, 2015, doi : 10.5445 / IR / 1000048047 .