Speaker: Ross Whitaker, Scientific Computing and Imaging Institute, University of Utah
Title:
Adaptive Markov Models with Information-Theoretic Methods for Image Analysis
Low-level problems in image processing, such as denoising,
reconstruction, and segmentation typically require some kind of model
of image structure. Thus, the modeling images in a general, yet
tractable manner remains an important area of research. Most image
processing algorithms make strong geometric or statistical assumptions
about the properties of the signal and/or noise. Therefore, they lack
the generality to be easily applied to new applications or diverse
image collections. This talk presents an adaptive Markov model of
images that allows algorithms to automatically learn the local
statistical dependencies of image neighborhoods. Probability
densities for neighborhoods are estimated nonparametrically, through a
kernel-based strategy, and thus, image statistics are captured through
large sets of examples of image neighborhoods. We use this strategy to
create adaptive algorithms for low-level image processing. We
incorporate prior information, when available, using optimal Bayesian
formulations. We enforce optimality criteria based on fundamental
information-theoretic concepts that capture the functional dependence,
information content, and uncertainty in the data. This talk presents
examples of the application of this strategy to denoising,
reconstruction, and segmentation.