Bayesian Robust Matrix Factorization for Image and Video Processing (ICCV2013')

Naiyan Wang and Dit-Yan Yeung.


Matrix factorization is a fundamental problem that is often encountered in many computer vision and machine learning tasks. In recent years, enhancing the robustness of matrix factorization methods has attracted much attention in the research community. To benefit from the strengths of full Bayesian treatment over point estimation, we propose here a full Bayesian approach to robust matrix factoriza-tion. For the generative process, the model parameters have conjugate priors and the likelihood (or noise model) takes the form of a Laplace mixture. For Bayesian inference, we devise an effiient sampling algorithm by exploiting a hierarchical view of the Laplace distribution. Besides the basic model, we also propose an extension which assumes that the outliers exhibit spatial or temporal proximity as encoun-tered in many computer vision applications. The proposed methods give competitive experimental results when compared with several state-of-the-art methods on some benchmark image and video processing tasks

We have updated the MATLAB codes. In the previous release, we provided a wrong version of the MBRMF code. We also improve the efficiency in this version. If you downloaded before, please retry it. Sorry for the inconvenience.

[pdf] [Supplemental Material] [Matlab Code] [BibTex]

Related Project

[PRMF Project Page]

Compare with State-of-the-art

The following figures show a synthetic experiment in the paper. In this task, we aim to remove a piece of randomly generated text from a low rank image. We compare our methods with six state-of-the-art methods in robust matrix factorization (a.k.a robust PCA) fields. More details and more experiments could be found in the paper!