Publication     Conference     Thesis

Robust Principal Component Analysis

Motivation
  Despite many years of research and development in computer vision and image processing, many difficult problems remain. These difficult problems include removal of reflection, shadow and specularity in images, modeling of background in videos, recovery of missing parts in images and videos, etc. Although various methods have been attempted to solve these problems, these methods tend to be specialised to a very narrow range of problems and they typically lack provable performance quality. Hence, knowledge and experience gained in solving one problem is not easily transferrable across application domains.

Recent research in robust principal component analysis (RPCA) offers a promising unified approach for solving the above problems. The RPCA approach makes use of information redundancy in multiple data samples and formulate the problem as one of decomposing a noise-corrupted data matrix into a low-rank matrix and a sparse noise or error matrix. The low-rank matrix contains noise-free versions of the input data samples. This approach has been demonstrated for solving, in a unified manner, a variety of problems such as removal of noise, shadow and specularity in images, modeling of background in videos, alignment of images, etc. Nevertheless, application of RPCA is so far limited to problems with small images and short videos. There are many obstacles to overcome before it can be applied to other visual data such as 3D mesh models and 3D motion data used in various visual computing application domains. These obstacles include non-sparse noise, non-linear relationships between data samples, large sample size, large data dimensionality, long temporal sequence, etc.
   
Research Goal
  Our research goal is to advance the research of low-rank matrix recovery for visual computing by overcoming the above obstacles so that a unified approach can be applied to solve a wide range of problems in various application domains.
   
Researchers
>> Dr. Lai Jian, Resarch Fellow, Dept. of Computer Science, Natioanl University of Singapore
>> Ms. Vaishali Sharma, Research Assistant, Dept. of Computer Science, National University of Singapore
>> Mr. Xie Shudong, Ph.D. student, Dept. of Computer Science, National University of Singapore
>> A/Prof. Leow Wee Kheng, Dept. of Computer Science, Natioanl University of Singapore
>> A/Prof. Terence Sim, Dept. of Computer Science, Natioanl University of Singapore
   
Publications
>> J. Lai, W. Leow, T. Sim, V. Sharma. Think Big, Solve Small: Scaling Up Robust PCA with Coupled Dictionaries. In Proc. IEEE Winter Conf. on Applications of Computer Vision, Mar 2016.
>> W. K. Leow, G. Li, J. Lai, T. Sim, V. Sharma. Hide and Seek: Uncovering Facial Occlusion with Variable-Threshold Robust PCA. In Proc. IEEE Winter Conf. on Applications of Computer Vision, Mar 2016.
>> Y. Cheng, S. Xie, W. K. Leow, K. Zhang. Plane-Fitting Robust Registration for Complex 3D Models. In Proc. Int. Conf. on Computer Analysis of Images and Patterns, Sep 2015.
>> J. Lai, W. K. Leow, T. Sim. Incremental Fixed-Rank Robust PCA for Video Background Recovery. In Proc. Int. Conf. on Computer Analysis of Images and Patterns, Sep 2015.
>> W. K. Leow, Y. Cheng, L. Zhang, T. Sim, L. Foo. Background Recovery by Fixed-rank Robust Principal Component Analysis. In Proc. Int. Conf. on Computer Analysis of Images and Patterns, Aug 2013.

3 July 2016