An orientation-adaptive Gaussian scale mixture model for image denoising

D K Hammond and E P Simoncelli

Published in SIAM Conf. on Imaging Science, May 2006.

This paper has been superseded by:
Image modeling and denoising with orientation-adapted Gaussian scale mixtures
D K Hammond and E P Simoncelli.
IEEE Trans. Image Processing, vol.17(11), pp. 2089--2101, Nov 2008.


We develop a model for patches of image wavelet coefficients that is explicitly adapted to local orientation. Image patches are described as samples of a Gaussian process that is rotated and scaled by hidden random variables representing the local image orientation and contrast, respectively. A Bayesian denoising method, based on conditioning on and integrating over the hidden variables, yields visually superior results when compared to previous scale mixture models that do not explicitly model orientation.
  • Listing of all publications