Image Compression with Diffusion

Prof. Joachim Weickert

Saarland University, Saarbruecken
Germany

Abstract:

The basic idea behind diffusion-based image compression is surprisingly simple: One only stores a small fraction of all pixels and inpaints the missing data with a suitable diffusion equation. While diffusion coding started as a proof-of-concept in 2005, it has improved enormously over the years. In the meantime it is on its way of becoming an interesting alternative to transform-based codecs such as JPEG and JPEG 2000.

The goal of this talk is to give an overview of the state-of-the-art in diffusion coding. In particular, we address the following questions:

• What is the best diffusion equation for data inpainting?
• Which pixels should be kept?
• How can the selected pixels be encoded in an efficient way?
• How good is the quality that can be achieved?
• Are such codecs fast enough for real-time applications?

Diffusion coding is not restricted to a single method: It is a versatile framework that can be tailored towards specific needs and applications. To demonstrate this, we will also have a look at progressive modes, codecs for shapes, cartoon-like images, colour images, surface data, 3-D data sets, depth images, and applications in steganography.

About the Keynote Speaker:

Joachim Weickert is professor of Mathematics and Computer Science at Saarland University (Saarbruecken, Germany), where he heads the Mathematical Image Analysis Group. He graduated and obtained his Ph.D. from the University of Kaiserslautern (Germany) in 1991 and 1996. He worked as post-doctoral researcher at the University Hospital of Utrecht (The Netherlands) and the University of Copenhagen (Denmark), and as assistant professor at the University of Mannheim (Germany).

Joachim Weickert has developed many models and efficient algorithms for image processing and computer vision using partial differential equations and optimisation principles. In particular he has contributed to diffusion filtering, motion analysis in videos, processing of tensor fields, and data compression. His scientific work covers more than 250 refereed publications. They have been cited more than 15000 times and have led to an h-index of 58. Joachim Weickert has served in the editorial boards of nine international journals and is editor-in-chief of the Journal of Mathematical Imaging and Vision. In 2010 he has received the Gottfried Wilhelm Leibniz Prize which is the highest German science award.