Fields of Experts:
A Framework for Learning Image Priors

Stefan Roth and Michael J. Black
IEEE Conference on Computer Vision and Pattern Recognition, volume 2, pages 860-867, San Diego, California, June 2005.

At CVPR 2005 we proposed a new model for learning the prior probability of generic images. The model is a higher-order Markov random field formulation, where the clique potentials are represented as a Product of Experts. The model parameters are learned using contrastive divergence from a database of generic, natural images. The model provides a probability density for full images and has direct applications to a variety of computer vision and image processing problems. In our paper we address image denoising and image inpainting in particular.

Quick links:

• Paper download [pdf, 493kB], [ps, 4.88MB], [ps.gz, 3.12MB]
• Denoising results
• Image inpainting results
• Training data: Sample images and download
• Test data: Sample images and download
• Example code and models