Biometric Writing Systems


Analysis of Hand Motor Function

The analysis of signatures, handwritten passwords, or hand drawings provides very interesting information about a person. It reflects her/his identity and handedness, it is influenced by temperature, alcohol, or physical strain, and it shows certain side-effects of drugs. Basically, it could be investigated either off-line data (e.g., scanned images) or on-line data (e.g., coordinate time series of a graphic tablet or force signals of specific pens). The latter offers more valuable information for certain tasks and is, therefore, addressed in our work.

Signature Verification and Identification

Biometric authentication is an important research issue since several years. Currently, there exist a broad range of techniques: some are very reliable but require some costly equipment (e.g., iris scan) and others are quite cheap but often not very precise (e.g., fingerprint). Techniques based on signatures of a person have the particular advantage that many people would easily accept this kind of authentication as they sign documents for that purpose almost daily, for example, whenever they pay with their credit card. Unfortunately, signature data exhibits some variability due to various influences. To increase the precision of a signature-based approach, we measure on-line data such as forces in three orthogonal directions, pen inclination, pen tip coordinates etc. over time and analyse the dynamics of a signature. We use standard graphic tablets (Wacom Intuos 3) as well as the BiSP (Biometric Smart Pen) developed at the University of Applied Sciences Regensburg.



Sample signature and the corresponding force signals in three orthogonal directions measured with the BiSP.

We focus on the classification of on-line signature data with support vector machines (SVM) equipped with a new kind of kernel function which was particularly designed to cope with the local and global variability of signature data. This kernel function is based on the computation of the length of the longest common subsequences (LCSS) of two time series. Our signature verification system ID-Signs achieves equal error rates of about 0.41% (proprietary data set with random forgeries) and 0.12% (random forgeries of the SVC 2004 data set). With these error rates it is one of the best-performing signature verification systems today. Instead of signatures, handwritten passwords can alternatively be used to recognize a person.



CeBit 2005: Dr. Christian Gruber shows our signature verification system ID-Signs to the Bavarian State Secretary Dr. Otto Wiesheu (Photo: Jörg Perwitzschky).

Currently, we develop a “signature safety checker” that can be used in a signature verification system to secure the enrolment process.

Biomedical Applications

We are also interested in biomedical applications of biometric writing systems such as the detection of side-effects of certain drugs or the development of handedness tests for preschool children. Basically, biometric writing systems could be used to address many other interesting problems in the field, e.g. to support a therapy for hyperactive children or a rehabilitation of stroke patients.

We made a first, but important step towards a quantification of side-effects of typical and atypical neuroleptics by comparing the fine motor skills of mentally diseased persons (schizophrenia) and healthy persons. In another study we addressed the influences of temperature, alcohol, and physical strain on the handwriting of healthy persons. In both cases we used the BiSP of the University of Applied Sciences, Regensburg.

Handedness is the preference of humans to use a dedicated hand for most manual tasks, e.g., writing with a pen or cutting with a knife. Handedness tests are needed for preschool children with uncertain handedness, for instances, to avoid a wrong writing education in school. We developed a handedness test based on graphic tablets that can be used by paediatricians or occupational therapists.



Flight subtest of the handedness test executed with the right and the left hand, respectively.



Further Information

Staff:

Publications:

C. Gruber, T. Gruber, S. Krinninger, B. Sick; Online Signature Verification with Support Vector Machines Based on LCSS Kernel Functions; in: IEEE Transactions on Systems, Man, and Cybernetics – Part B (Cybernetics); vol. 40, no. 4, pp. 1088-1100; 2010


C. Gruber, B. Sick; A Comparison of Biometric Writing Systems for the Analysis of Human Fine Motor Skills; in: Proceedings of the ”IEEE Three-Rivers Workshop on Soft Computing in Industrial Applications (SMCia/07)”; pp. 49-54; Passau, 2007


M. Dose, C. Gruber, A. Grunz, C. Hook, J. Kempf, G. Scharfenberg, B. Sick; Towards an Automated Analysis of Neuroleptics’ Impact on Human Hand Motor Skills; in: Proceedings of the ”2007 IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB 2007)”; pp. 494-501; Honolulu, 2007


J. Hofer, C. Gruber, B. Sick; Biometric Analysis of Handwriting Dynamics Using a Script Generator Model; in: Proceedings of the ”2006 IEEE Mountain Workshop on Adaptive and Learning Systems (SMCals/06)”; pp. 36-41; Logan, 2006


T. Gruber, C. Gruber, B. Sick; Online Signature Verification With new Time Series Kernels for Support Vector Machines; in: D. Zhang, A. K. Jain (Eds.): Advances in Biometrics: International Conference ICB 2006; Lecture Notes in Computer Science 3832, Springer Verlag, Berlin, Heidelberg, New York; pp. 500-508; Hong Kong, 2006


and others...

Awards:

  • Fresenius Awards 2006 Inventors’ Prize (C. Gruber, C. Hook, J. Kempf, G. Scharfenberg, and B. Sick) at MEDICA 2006

Collaboration: