Suche
News
Sie sind hier
Computer Vision
The Computer Vision Group conducts research in the field of mathematical image processing, computer vision, and machine learning with a special focus on ways to combine machine learning and energy minimization methods. In this context we are interested in efficient optimization techniques, convex relaxations, inverse problems, and provable properties of deep neural networks in the aforementioned areas.
Head of Chair
Prof. Dr. Michael Moeller
email: michael.moeller@uni-siegen.de
Web page: Computer Vision
The Confluence of Model- and Learning-based approaches
While learning based approaches have shown to be extremely powerful for various image reconstruction and computer vision tasks, the usage of plain forward networks may lead to unexpected results, event to safety risks, when applied with any further control mechanism. We are working on introducing model-based information to neural networks in order to make networks more robust, particularly in situations of little training data.
Efficient and Near-global Optimization Methods
Most approaches to computer vision and image reconstruction problems result in high dimensional optimization problems, including the training of neural networks themselves. We are studying efficient numerical algorithms to solve these problems, specifically focusing on how 'good' solutions can be computed even for non-convex problems, for instance through convex relaxation methods.