New system for analysing handwriting

Correct handwriting identification can be very important in a court case to determine who is the writer of, for example, a handwritten threatening letter or a possibly forged suicide note. Handwriting biometrics is the technology where computers analyse handwriting. Axel Brink has developed a new computer technology (Quill) to verify and identify handwriting. He will be awarded a PhD for his research by the University of Groningen on 2 December 2011.

Handwriting biometrics is the attribution of an author to a handwritten text by having a computer analyse characteristics of the handwriting. Handwriting biometrics can be used both in author verification and in author identification. An author verification system compares the handwriting in a document with that in a different document. An author identification system searches in a collection on the basis of one text for documents in the same handwriting.

Quill

Axel Brink has developed a new technology for handwriting biometrics. This technology, called Quill, measures the precise direction and width of the ink trace for every pixel on the edge of a letter or digit. It lists thousands of measurement points on a single page of text. Together, all these measurement points form the handwriting profile. That profile can be used in author verification and identification. 

Medieval

Quill is as efficient in the analysis of modern and medieval handwriting as the best handwriting biometrics methods Brink researched. Together with other technologies, Quill has been included in GIWIS (Groningen Intelligent Writer Identification System), a programme that can be used for criminal investigation and historical research.

Robust

The PhD student also investigated the robustness of a number of other handwriting biometrics methods. Brink: ‘Robust in this context means that the technology is able to identify the correct author from a large collection of examples of handwriting from actual practice. A number of existing technologies perform well with carefully written texts, but that doesn’t say much about their performance under realistic conditions. Often words in the examples are crossed out or are not written in straight lines, and the paper of medieval manuscripts is often distractingly discoloured.’

Slant

One important aspect that Brink investigated is handwriting being disguised by slant – making the characters lean more towards the left or the right. Slant is a simple and popular way to disguise handwriting – it immediately looks different. With the help of a computer you can reverse this type of disguise to a certain degree.
Brink: ‘I examined the recognizability of disguised handwriting and could show that after correction it was less successfully identified then “natural” handwriting. Computers can apparently only partially reverse the effects of that kind of disguise.’ This is probably because people not only adapt the angle of the slant, but unconsciously also change other characteristics of the handwriting. So, with regard to disguised handwriting, such systems are not yet robust.

A hundred characters

Brink’s robustness research also revealed that you need about a hundred characters for successful author verification and identification. Brink: ‘The more characters there are available for analysis, the more precisely we can compare handwriting. It turns out that the results do not improve all that much with more than a hundred characters. However, fewer than a hundred characters does mean a less good result.’
The third result of Brink’s robustness research is that crossed-out words can remain in a text without affecting the quality of the analysis.

Curriculum Vitae

Axel Brink (Dalfsen, 1979) studied computer science at the University of Groningen. He also gained a propaedeutic certificate in communication and information sciences. In December 2005, Brink started his PhD research at the Institute of Artificial Intelligence and Cognitive Engineering (ALICE) at the University of Groningen. His supervisor was Prof. L.R.B. Schomaker and his thesis is entitled Robust and applicable handwriting biometrics. Brink will be awarded his PhD on 2 December 2011. Since January 2010 he has been a knowledge analyst for Be Informed in Apeldoorn.

Note for the press

Contact: Axel Brink; e-mail axelbrink gmail.com