Image Understanding And Pattern Recognition

IUPR research focuses on large scale, practical applications of pattern recognition, image understanding, machine learning, and statistical language modeling. The fundamental techniques we are developing are:* practical learning algorithms that scale to very large problem sizes * automatic parameter selection and validation for pattern recognition systems * problem instance-specific adaptation * validation and benchmarking of systems * massively parallel methods * statistical analysis of text and images Optical character recognition is an excellent area for studying and evaluating approaches to these problems, as well as a useful application. We are currently building a system called OCRopus, designed for large scale digital library applications. A second application is high quality 3D capture of books and art using low-cost digital cameras as part of project DECAPOD. As part of project PaREn, we are creating datasets and tools intended to improve the evaluation and validation of pattern recognition methods. The MOVIMOS and Moonvid projects are creating a next-generation content-based image retrieval system, also with applications to forensics and security. IUPR is funded through industrial contract research and public research grants (EU, federal, state).