IC Monday Seminar : Image Compression with Differential Equations
Event details
| Date | 03.10.2011 |
| Hour | 16:15 |
| Speaker | Prof. Joachim Weickert - Saarland University, Germany - Invited by Prof. Pascal Fua |
| Location |
INM 202
|
| Category | Conferences - Seminars |
Abstract One of the most fascinating aspects of partial differential equations (PDEs) in image processing is their filling-in effect. It allows to recover missing image structures by interpolating data from the neighbourhood. In this talk we exploit this property for lossy compression of digital images. The idea is to keep only a small fraction of the pixels and to reconstruct the remaining data with PDE-based interpolation. This gives rise to three interdependent questions: 1. Which are the best pixels for being kept? 2. How can the selected pixels be encoded in an efficient way? 3. What are the most useful PDEs for data interpolation? In order to gain insights into the pixel selection problem we first focus on interpolation by homogeneous diffusion, where we present theoretical results, optimization strategies and fast algorithms. To encode the selected data efficiently, we use contour data or tree-based representations. Afterwards we evaluate various PDE-based interpolation operators and show why anisotropic nonlinear diffusion has very favourable properties. It is used to design a codec that can outperform the JPEG standard and even its more advanced wavelet-based successor JPEG 2000.
Links
Practical information
- General public
- Free
Contact
- Simone Muller