Bayesian dictionary learning with Gaussian processes and sigmoid belief networks


Conference Paper

In dictionary learning for analysis of images, spatial correlation from extracted patches can be leveraged to improve characterization power. We propose a Bayesian framework for dictionary learning, with spatial location dependencies captured by imposing a multiplicative Gaussian process (GP) priors on the latent units representing binary activations. Data augmentation and Kronecker methods allow for efficient Markov chain Monte Carlo sampling. We further extend the model with Sigmoid Belief Networks (SBNs), linking the GPs to the top-layer latent binary units of the SBN, capturing inter-dictionary dependencies while also yielding computational savings. Applications to image denoising, inpainting and depth-information restoration demonstrate that the proposed model outperforms other leading Bayesian dictionary learning approaches.

Duke Authors

Cited Authors

  • Yizhe, Z; Henao, R; Li, C; Carin, L

Published Date

  • January 1, 2016

Published In

Volume / Issue

  • 2016-January /

Start / End Page

  • 2364 - 2370

International Standard Serial Number (ISSN)

  • 1045-0823

Citation Source

  • Scopus