Project

Back to overview

Automatic Handwriting Recognition and Writer Identification based on the Kinematic Theory

Applicant Fischer Andreas
Number 151279
Funding scheme Advanced Postdoc.Mobility
Research institution Département de Génie Electrique Ecole Polytechnique de Montréal
Institution of higher education Institution abroad - IACH
Main discipline Mathematics
Start/End 01.04.2014 - 28.02.2015
Show all

Keywords (7)

Handwriting Recognition; Artificial Intelligence; Pattern Recognition; Kinematic Theory; Writer Identification; Synthetic Handwriting; Neuromuscular Handwriting Representation

Lay Summary (German)

Lead
Wir kommunizieren nach wie vor häufig mit unserer Handschrift, trotz zahlreichen technologischen Erneuerungen wie der Erfindung der Druckmaschine, der Schreibmaschine, des Personal Computers und des tragbaren Computers. Um diese praktische und persönliche Art uns auszudrücken in die digitale Welt zu integrieren, braucht es Methoden der Künstlichen Intelligenz und der Mustererkennung, hauptsächlich um die Handschrift automatisch in Maschinenschrift zu überführen und um den Schreiber zu identifizieren, beispielsweise für die biometrische Authentifizierung oder für forensische Untersuchungen. In diesem Projekt untersuchen wir die Handschrifterkennung aus einer neuen, kinematischen Perspektive, welche den neuromuskulären Schreibprozess ins Zentrum stellt.
Lay summary

Inhalt und Ziel des Forschungsprojekts

Im Kontext der kinematischen Theorie schneller Bewegungen wurde kürzlich ein mathematisches Modell des neuromuskulären Schreibprozesses vorgestellt, dessen Modellparameter automatisch vom Bewegungsverlauf des Schreibstifts abgeschätzt werden können. Unser Ziel ist es, dieses Schreibmodell für die Handschrifterkennung einzusetzen. Einerseits entwickeln wir Methoden für die künstliche Erzeugung von Handschriften, um eine grosse Datenbank von Lernbeispielen aufzubauen, welche die Genauigkeit bestehender Erkennungssysteme verbessern kann. Andererseits entwickeln wir ein System für die Schreiberidentifikation, welches auf biometrischen Merkmalen des Schreibprozesses beruht.

Wissenschaftlicher und gesellschaftlicher Kontext des Forschungsprojekts

Direkte Anwendungen unserer Forschung liegen im Bereich der maschinenschriftlichen Transkription, biometrischen Authentifizierung und forensischen Untersuchung von Handschriften. Darüber hinaus sehen wir ein Potential für biomedizinische Anwendungen. Die biometrischen Merkmale des Schreibprozesses, welche wir für die Schreiberidentifikation entwickeln, könnten interessant sein für Diagnose- und Prognosemethoden von Krankheiten, die mit der Bewegung zusammenhängen, so wie beispielsweise die Parkinson-Krankheit. Schliesslich könnte das Schreibmodell auch dafür eingesetzt werden, um Kinder mit einer interaktiven Lernhilfe dabei zu unterstützen, das Schreiben zu lernen.


Direct link to Lay Summary Last update: 05.02.2014

Responsible applicant and co-applicants

Publications

Publication
A Dissimilarity Measure for On-Line Signature Verification Based on the Sigma-Lognormal Model
Fischer A., Plamondon R. (2015), A Dissimilarity Measure for On-Line Signature Verification Based on the Sigma-Lognormal Model, in Proc. 17th Conf. of the International Graphonomics Society, Université des Antilles, Schoelcher cedex, Martinique.
Approximation of graph edit distance based on Hausdorff matching
Fischer A., Suen C. Y., Frinken V., Riesen K., Bunke H. (2015), Approximation of graph edit distance based on Hausdorff matching, in Pattern Recognition, 48(2), 331-343.
Building Classifier Ensembles Using Greedy Graph Edit Distance
Riesen Kaspar, Ferrer Miquel, Fischer Andreas (2015), Building Classifier Ensembles Using Greedy Graph Edit Distance, in Proc. 12th Int. Workshop on Multiple Classifier Systems, Springer, Cham.
Estimating Graph Edit Distance Using Lower and Upper Bounds of Bipartite Approximations
Riesen K., Fischer A., Bunke H. (2015), Estimating Graph Edit Distance Using Lower and Upper Bounds of Bipartite Approximations, in Int. Journal of Pattern Recognition and Artificial Intelligence, 29(2), 1-27.
Improving Hausdorff edit distance using structural node context
Fischer A., Uchida S., Frinken V., Riesen K., Bunke H. (2015), Improving Hausdorff edit distance using structural node context, in Proc. 10th Int. Workshop on Graph-based Representations in Pattern Recognition, Springer, Berlin, Heidelberg.
Omega-Lognormal Analysis of Oscillatory Movements as a Function of Brain Stroke Risk Factors
Bou Hernandez A., Fischer A., Plamondon R. (2015), Omega-Lognormal Analysis of Oscillatory Movements as a Function of Brain Stroke Risk Factors, in Proc. 17th Conf. of the International Graphonomics Society, Université des Antilles, Schoelcher cedex, Martinique.
Robust score normalization for DTW-based on-line signature verification
Fischer Andreas, Diaz Moises, Plamondon Réjean, Ferrer Miguel A. (2015), Robust score normalization for DTW-based on-line signature verification, in Proc. 13th Int. Conf. on Document Analysis and Recognition, IEEE Computer Society Press, Los Alamitos, CA.
Towards an automatic on-line signature verifier using only one reference per signer
Diaz Moises, Fischer Andreas, Plamondon Réjean, Ferrer Miguel A. (2015), Towards an automatic on-line signature verifier using only one reference per signer, in Proc. 13th Int. Conf. on Document Analysis and Recognition, IEEE Computer Society Press, Los Alamitos, CA.
A cache language model for whole document handwriting recognition
Frinken V., Karatzas D., Fischer A. (2014), A cache language model for whole document handwriting recognition, in Proc. 11th Int. Workshop on Document Analysis Systems, IEEE Computer Society Press, Los Alamitos, CA.
A Combined System for Text Line Extraction and Handwriting Recognition in Historical Documents
Fischer A., Baechler M., Garz A., Liwicki M., Ingold R. (2014), A Combined System for Text Line Extraction and Handwriting Recognition in Historical Documents, in Proc. 11th Int. Workshop on Document Analysis Systems, IEEE Computer Society Press, Los Alamitos, CA.
A feature extraction method for cursive character recognition using higher-order singular value decomposition
Ameri M. R., Haji M., Fischer A., Ponson D., Bui T. D. (2014), A feature extraction method for cursive character recognition using higher-order singular value decomposition, in Proc. 14th Int. Conf. on Frontiers in Handwriting Recognition, IEEE Computer Society Press, Los Alamitos, CA.
A Hausdorff Heuristic for Efficient Computation of Graph Edit Distance
Fischer A., Plamondon R., Savaria Y., Riesen K., Bunke H. (2014), A Hausdorff Heuristic for Efficient Computation of Graph Edit Distance, in Proc. Int. Workshop on Structural, Syntactic, and Statistical Pattern Recognition, Springer, Berlin, Heidelberg.
Combining bipartite graph matching and beam search for graph edit distance approximation
Riesen K., Fischer A., Bunke H. (2014), Combining bipartite graph matching and beam search for graph edit distance approximation, in Proc. 6th Int. Workshop on Artificial Neural Networks in Pattern Recognition, Springer, Berlin, Heidelberg.
Computing upper and lower bounds of graph edit distance in cubic time
Riesen K., Fischer A., Bunke H. (2014), Computing upper and lower bounds of graph edit distance in cubic time, in Proc. 6th Int. Workshop on Artificial Neural Networks in Pattern Recognition, Springer, Berlin, Heidelberg.
Improving Approximate Graph Edit Distance Using Genetic Algorithms
Riesen K., Fischer A., Bunke H. (2014), Improving Approximate Graph Edit Distance Using Genetic Algorithms, in Proc. Int. Workshop on Structural, Syntactic, and Statistical Pattern Recognition, Springer, Berlin, Heidelberg.
Improving graph edit distance approximation by centrality measures
Riesen K., Fischer A., Bunke H. (2014), Improving graph edit distance approximation by centrality measures, in Proc. 22nd Int. Conf. on Pattern Recognition, IEEE Computer Society Press, Los Alamitos, CA.
Neuromuscular representation and synthetic generation of handwritten whiteboard notes
Fischer A., Plamondon R., O'Reilly C., Savaria Y. (2014), Neuromuscular representation and synthetic generation of handwritten whiteboard notes, in Proc. 14th Int. Conf. on Frontiers in Handwriting Recognition, IEEE Computer Society Press, Los Alamitos, CA.
Dynamic Signature Verification System Based on One Real Signature
Diaz Moises, Fischer Andreas, Ferrer Miguel A., Plamondon Réjean, Dynamic Signature Verification System Based on One Real Signature, in IEEE Trans. on Cybernetics, PP(99), 1-12.
Signature Verification Based on the Kinematic Theory of Rapid Human Movements
Fischer Andreas, Plamondon Réjean, Signature Verification Based on the Kinematic Theory of Rapid Human Movements, in IEEE Trans. on Human-Machine Systems, PP(99), 1-12.

Collaboration

Group / person Country
Types of collaboration
Seiichi Uchida, Human Interface Laboratory, Kyushu University Japan (Asia)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
Ching Y. Suen, CENPARMI, Concordia University Canada (North America)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
Miguel A. Ferrer, GPDS research group, University of Las Palmas de Gran Canaria Spain (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
Rolf Ingold, DIVA research group, University of Fribourg Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
Kaspar Riesen, professor at the University of Applied Sciences and Arts Northwestern Switzerland Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication

Associated projects

Number Title Start Funding scheme
141453 Bootstrapping Handwriting Recognition Systems for Historical Documents 01.10.2012 Fellowships for prospective researchers

Abstract

Handwriting as a means of communication has survived many technological revolutions including the invention of the printing press, typewriter, personal computer, and portable computer. In order to incorporate this convenient and personal way to express ourselves into the digital world, artificial intelligence and pattern recognition are needed, most prominently to transcribe handwriting into computer-readable text and to distinguish different writers in the context of biometric authentication and forensic analysis. Although significant progress has been made in research over the past decades and accurate commercial systems have been developed for constrained scenarios, these challenging problems are still far from being solved.In this project, we approach handwriting recognition from a novel promising direction, that is from a kinematic point of view. Recent progress in this domain includes the derivation of a mathematical model of the neuromuscular process that generates complex pen movements, alongside with the means to extract the model parameters from sample handwritings. We believe that the integration of this knowledge about the writing process into automatic handwriting recognition systems could be highly beneficial.Our first goal is the development of kinematic methods for synthetic handwriting generation. The idea is to modify existing handwriting samples by introducing some variance in the pen movement. This allows us to automatically generate a large database of learning samples for training handwriting recognition systems and improving their accuracy. When compared with geometric distortion models, the neuromuscular model is expected to generate more natural handwriting samples that are well suited as learning samples.Secondly, we apply our kinematic approach to the task of writer identification. This technology allows personalized handwriting recognition and has an important application in forensics to support human experts in their work, for instance by narrowing down lists of candidates that could have written a document in question. We aim to develop novel identification methods based on the behavioral biometrics provided by the neuromuscular handwriting model.Future applications of the research conducted in this project include biomedical engineering. In particular, the neuromuscular movement characteristics that we develop for writer identification could be instrumental to devise diagnosis and prognosis methods for movement-related medical conditions such as Parkinson's disease. Last but not least the kinematic model of handwritten words and sentences could be interesting for the development of teaching aids that help children learn how to write.
-