Statistical Models for Images

Research workshop in Luminy, 17-21 May 2010



Titles and abstracts


 Practical informations 

 workshop 2008 


       Image processing has connections with many domains of applied mathematics: variational methods, harmonic analysis, partial differential equations, stochastic models, etc. These last years, the importance of statistical methods has constantly increased, and the interaction between researchers in image processing and statistics is needed on many subjects.
       Variational methods, for example, are generally designed using a-priori differential operators, but they can be often interpreted in a Bayesian framework, which yields interesting perspectives to improve models from objective observations on images.
       More generally, the multiplication of data (images, pixels) increases the need for new models capable of extracting image properties at a large scale. Statistics are a very natural way to handle these huge amounts of data.
       Last, precisely because of the multiplication of data sources, the need for non-supervised algorithms has increased in image analysis. Again, the probabilistic/statistical framework is well adapted to this task, since it offers a natural modeling of decision processes.

       The aim of this workshop is to favor exchanges between image processing specialists motivated by these statistical issues, and statisticians interested in applications to image processing. Among themes that may be addressed during this workshop, are:
  • Statistics of natural images
  • Probabilistic models for medical images and textures
  • Denoising and image restoration
  • Structure detection in images
  • Compositional models
  • Parametric and non-parametric estimation
  • Model selection
  • Intrinsic dimension
  • Data mining
  • etc.


Agnès Desolneux, Lionel Moisan, Frédéric Richard, from MAP5 (University of Paris Descartes)

Last update: September 18, 2009