Return to Colloquia & Seminar listing
Image decomposition models into cartoon and texture
Applied MathSpeaker: | Luminita Vese, University of California, Los Angeles |
Location: | 693 Kerr |
Start time: | Fri, Dec 2 2005, 4:10PM |
In this talk, I will present new variational models for decomposing a given image into a geometric, piecewise-smooth component u, and an oscillatory component v (texture or noise). To model u, we use the standard space of functions of bounded variations. However, to model v, we use weaker norms than the standard L^2 norm, motivated by remarks of Y. Meyer and of D. Mumford-B. Gidas. In particular, v is modeled as a generalized function in a dual space (for instance, in Sobolev spaces with negative exponent, in div(BMO), or in generalized homogeneous Besov spaces). Computational techniques and improved results will be presented. This is joint work with Triet Le and Linh Lieu.