糖心TV

The tremendous growth of digital multimedia content in the past few years has led to the growing concern and requirement for copyright protection for proof of ownership and content authentication. In this talk I will give an overview on digital watermarking and authentication of images and highlight the differences between robust, fragile and semi-fragile watermarking for law enforcement applications such as crime scene analysis and traffic enforcement.  Another research area which is gaining momentum is binary image watermarking and some current work will also be presented.

Bio:

Professor Anthony TS Ho joined the Department of Computing, Faculty of Engineering and Physical Sciences, University of Surrey in January 2006 and holds a Personal Chair in Multimedia Security. He was an Associate Professor at Nanyang Technological University (NTU), Singapore from 1994 to 2005. He has been working on digital watermarking and steganography since 1997 and co-founded DataMark Technologies in 1998, one of the first spin-out companies by an academic at NTU and one of the first companies in the Asia-Pacific region, specializing in the research and commercialization of digital watermarking technologies. He leads and coordinates the research activities of a large research group on Watermarking and Multimedia Security in the Department of Computing. Professor Ho won the prestigious IET Innovation in Engineering Award under the security category for his research in digital watermarking in 2006.  

"The Morphon: A Robust Tool for Image Fusion and Segmentation"

Hans Knutsson, Linköpings Universitet, Sweden 

Even though the history of image segmentation goes back to the very beginning of image processing it is still a topic of major concern. Naive thresholding based approaches are basically abandoned and the need to incorporate strong prior information is hardly disputed. A-priori information in terms of allowable deformation of initial shapes has been frequently used in segmentation of 2D images and for tracking of deformable objects in video.

Existing deformable models are however sensitive to the initial conditions of the computation and often get trapped in local minima. In such cases interactive measures are paramount to attain successful results. For most users interaction with the deformable model is complicated and non-intuitive and there is typically no means for user interaction during the process.

The talk targets segmentation and, to some extent, fusion of medical data volumes. The approach is rather general and encompasses much of the deformable model ideas that have evolved. Looking to achieve robustness as well as flexibility the approach aims for easy interactive incorporation of domain specific knowledge. The idea is to use a type of 'paint on priors' interface to specify a Morphon model for a specific situation. This model volume describes the pertinent general characteristics, such as shape, object elasticity, visual appearance etc, of the object. These priors determine how the neighborhood will be perceived and how this percept translates into a suggested new location. The model volume can be seen as an elastic (N-dimensional) canvas which is deformed/morphed to align with the data at hand, hence the method name - The Morphon.

Initial tests in several different cases show promising results. Examples will be given of heart wall segmentation in ultrasound sequences and extraction of brain nerve fibre tracks. The latter involves a novel approach where the displacement field is computed through the inverse gradient of a a rotational field holding the local difference in model and data orientation.