My general interests are pattern recognition, machine learning and computer vision. My current work at Siemens Corporate Research is concerned with object detection in automotive, medical, and aerial images. In my Ph.D. work, I have been researching within the on-line handwriting recognition system frog on hand.

Publications

Claus Bahlmann, Gunther Heidemann and Helge Ritter. Artificial Neural Networks for Automated Quality Control of Textile Seams. In Pattern Recognition, volume 32, number 6, June 1999.

Claus Bahlmann. Künstliche Neuronale Netze zur optischen Qualitätskontrolle textiler Nähte. M.S. thesis, Computer Science Department, University of Bielefeld, 1997 [in German].

Claus Bahlmann, Marc Schael and Hans Burkhardt. Aktives Sehen. Manual Student's computer vision laboratory, Computer Science Department, Albert Ludwigs University Freiburg, 2000 [in German].

Marc Schael, Claus Bahlmann and Hans Burkhardt. Klassifikatorentwurf. Manual Student's computer vision laboratory, Computer Science Department, Albert Ludwigs University Freiburg, 2000 [in German].

Some of my students' theses with HWR related work

Honors

Awarded the "Wolfgang-Gentner-Nachwuchsförderpreis" 2005 for the Ph. D. dissertation

Ph. D. work entitled Advanced Sequence Classification Techniques Applied to Online Handwriting Recognition honored "mit Auszeichnung" (equivalent to "with highest honors", "summa cum laude") in 2005.

"Best Paper Presentation" at Int. Workshop on Frontiers in Handwriting Recognition (IWFHR) 2002, Niagara-on-the-Lake, Canada, August 2002, for the contribution On-line Handwriting Recognition using Support Vector Machines - A kernel approach.

Co-Author "Best Student Paper" at IEEE Intelligent Vehicles Symposium (IV 2008), Eindhoven, The Netherlands, June 2008, for the contribution Real-Time Recognition of U.S. Speed Signs.

[back to top]

Teaching

• Assistant lecturer for the lecture Digitale Bildverarbeitung (digital image processing).
• Assistant lecturer for the lecture Mustererkennung (pattern recognition).

• Tutor for the computer vision practical laboratory Aktives Sehen (active vision), Mustererkennung (pattern recognition) and Klassifikatorentwurf (classifier design).
• Tutor for the practical laboratory Softwarpraktikum (software laboratory).
• Tutor of various B.S. and M.S. theses in the field of handwriting recognition

[back to top]

Past Projects

Handwriting Recognition (HWR)

"frog on hand" is the the freiburg recognition of on-line handwriting. Our goal is the design of a writer independent on-line handwriting recognition system.

In this work, a system including pre-processing, feature selection and two complementary classification approaches has been developed. Best results on the international UNIPEN benchmark have been achieved.

A special focus in this work is the classification. The classification approaches are based on two complementary methods: the so-called CSDTW (cluster generative statistical dynamic time warping) technique and support vector machines (SVMs).

A real-world application of our research is an implementation of recognition system on a Linux Compaq iPAQ.

[>> more information on my frog on hand page...]

[back to top]

Artificial Neural Networks for Automated Quality Control of Textile Seams - M.S. thesis

In this project I was working on a system for an automated, vision based quality control for textile seams. The developed system consists of

1. a suitable acquisition setup,
2. a seam detection stage, which very reliably transforms obliquely acquired seam images to a normalized position
3. a feature extraction stage, which is based on selected Fourier coefficients of one-dimensional image columns and
4. a self-organizing feature map (neural network) for classification.

The performance of the system has been evaluated by the classification of seam specimens, which are used in industrial textile manufacturing for the setting of the sewing machine parameters. The results have shown that even with few, but well-fashioned features good classification results can be obtained. The classification rate amounts to 80% correct classifications, the rest differs from the correct grade only by one (on a scale of five). We have shown that this result is not worse than the error of human experts, which can be measured by the disagreement among a set of different expert judgments.

Time needed for classification is about one second on a 130 MHz PC, which is much less than textile experts need for classification (30 seconds).

For more specific information please refer to my M.S. thesis or our Pattern Recognition article.

[back to top]