The explosive growth of digital content and of new multimedia applications expanded over heterogeneous networks has produced a pressing need for data representations that simultaneously support high-quality scalable image and video compression as well as image and video analysis tasks, such as denoising. New trends in image processing are geometry adapted sparse representations (curvelets, contourlets, bandelets, ...), which move the current limits of the visual quality achieved with classical bases, such as wavelet bases. While initial results on still images are undoubtedly impressive, the full potential that these new geometric image representations offer for denoising and compression is still far from being exploited. Additionally, beyond still images, algorithms for moving pictures (video) that are based on such content-adapted (geometric) sparse representations do not exist yet.
The aims of this project are to design representations adapted to the spatio-temporal geometry of the video sequence, and to develop advanced methods for denoising and scalable coding that are based on such representations. Additionally, an especially challenging topic will be the theoretical and practical development of jointly optimised denoising and compression techniques. For efficiently adapting the algorithms to the spatio-temporal context we will combine statistical context models with a fuzzy logical methodology. We believe that the fuzzy logical methodology will enable a true soft adaptation to the local context in real time as well as the proper treatment of ambiguities in motion estimation.

Runtime: 2006 - 2007