My Kind of Research
More information on my current research projects can be found on the Vision & Interaction Group (vintage) site.
Lossy compression methods for visual information in digital form
introduce distortions whose perceptibility highly depends on scene
content. Measuring the subjective visibility of these artifacts
accurately and reliably is difficult. The focus of my research is on
metrics for video quality assessment.
I have also compiled a page with useful resources for image and video quality assessment.
The ease of reproduction and distribution of images in digital form has increasingly become a concern for their creators. Therefore, the protection of authors' rights grows more and more important. Besides, it can be of interest to follow the distribution of such data. One solution to these problems is digital watermarking. This concept is based on the insertion of information into the data in such a way that the added information is not visible and yet resistant to image alterations.
Machines need the ability to determine the pose of objects in their environment in order to be able to reliably and intelligently interact with them. We investigated the neural network approach to model-based object pose estimation. Kohonen maps and some of their variations were studied for this purpose. We showed that the performance of these networks depends heavily on the mathematical representation of pose in general and orientation in particular. It is also advantageous to choose the topology of the neural network in harmony with the representation used.
The spatial interpretation of image features is substantially simplified if both the internal parameters of the camera mapping and the camera pose with respect to a task-specific reference frame are known. I participated in the development of a calibration procedure that automatically identifies both internal and external parameters of a stereo camera mounted on robot gripper. For such systems, not only the relative camera orientation has to be determined, but also the pose of the cameras with respect to the gripper-tip (hand-eye calibration). The method does not only account for a variety of typical lens distortions, but also for inaccuracies in robot kinematics.