What I do

I develop intelligent pattern recognition systems for image and multimedia analysis. How can we build intelligent computer systems that help us to extract information out of image and video collections ? How to organize the information items in a suitable form (e.g. into an non-classical data bank index) so that we can find efficiently what we are looking for ? This problem is not yet solved but we are making good progress in developing more and more intelligent machines.

As a software engineer, developer and reseacher,  I have specialized in image and multimedia analysis, computer vision and video analysis.  Currently, as a freelancer and founder of my own small consulting company EckesVision IT-Systems  I am realizing some small innovated projects, some in close colaboration with Trackmen and WendWerk, while preparing some larger ones.

 I would be happy to help you in realizing your ideas or to make a good system even better -- let's do it together, do not hestitate, contact me ...   

The central theme: Fusion of modalities, sensors and knowledge

There is a recurrent theme in the many projects I have worked on: Scientifically, one big problem in multimedia pattern recognition is the integration of different information sources: How to  integrate multiple cues,  for instance in face, person and object recognition, classification and segmentation tasks, is unclear and still subject to many research papers. My own work in face, person and object recognition has used low-level cues such as color, motion, stereo or texture but also high level cues like prior knowledge stemming from object recognition processes (e.g. the usage of cluttered scenes recognition results in my diploma thesis or in the WCNN paper) or other classification systems (e.g. usage of SMM/I ice concentration and weather information within the EU project ICEROUTES ). The proposed integration method is rather dynamic by letting the information items interact which each other: My research in image segmentation is based on multi scale pyramids of POTTS spins interacting with each other (simple MRF model). The interaction is derived by real-space renormalization methods and the resulting optimization problem is solved by a coarse-to-fine relaxation dynamics. The model is an abstract caricature of the neural segmentation mechanism present in the brain which is based on neural synchronization and de-synchronization of layers of spiking neurons. My work in object recognition is based on Elastic Graph Matching with graphs labeled with different features, such as color, texture and shape properties which are combined within a weak fusion architecture. Part of this work has already been patented and is published in a leading journal. .

At the Fraunhofer IAIS , I have extended these techniques to multimedia indexing, both offline but also online as part of the research project LIVE http://www.live.org. Topics include mutimodal face and person recognition, filtering algorithms and person/hand gesture recognition for Man-Machine Interfaces based on Time-Of-Flight cameras.

It is my strong believe that such research must be tested as part of real world applications: Hence,  I have developed  Motion Capturing plugins for Advanced Realtime Tracking tracking hardware to process and visualized human motion in AUTODESK products. I pitched a project to sense and count pedestrians for enhanced digital signage and worked as a project manager and software developer on face and person recognition in order to make organizing and sorting of personal photo collections much simpler -- in close cooperation with a major german multimedia software publisher. I also pitched and co-developed the Frauhofer IAIS iFinder -- a MPEG-7 based multimedia indexer for video speeches taken from the german parlament which is now also used in other applications. 

Besides this problem-orientated work in pattern recognition, I am also interested in good software engineering and OO-design since software. I supported actively the C++-library FLAVOR -- a framework for elastic graph matching for face and object recognition developed at the Institut für Neuroinformatik. At Fraunhofer I lead the development team of iFinderSDK -- a C++ framework for multimedia indexing. Moreover, I perform project management in research as well as industry projects.