伍世虔教授:Adaptive Robust Principal Component Analysis: Principle and Applications in Computer Vision
发布时间: 2019-11-26 浏览次数: 4581

报告内容:Adaptive Robust Principal Component Analysis: Principle and Applications in Computer Vision

报告人:伍世虔 教授

报告时间:11月28日 9:00

报告地点:现代交通工程中心7950会议室

  

报告人简介

伍世虔南洋理工大学博士,湖北省“楚天学者”特聘教授,IEEE高级会员,中国高被引学者(2014-2018,Elsevier)。现任武汉科技大学特聘教授博士生导师,智能信息处理与实时工业系统湖北省重点实验室主任,机器人与智能系统研究院副院长。曾任华中科技大学机械学院先进制造技术研究所副所长,新加坡国家科技局高级研究员(科学家),多次担任国际会议(ICICS、ICIEA、ISITC、ICoIAS等大会主席程序委员会主席或分会主席参与科研项目18项,其中主持11项,包括新加坡国家科技局、国家自然科学基金等课题。出版专著二本在国内外顶级期刊或会议发表学术论文200余篇,其中热点文章2篇,高被引文章9篇,总引用超过5000次主要研究方向有机器视觉、模式识别、机器学习及智能机器人等。


 

报告内容简介:

Robust Principal Component Analysis (RPCA) by decomposing a matrix into low-rank plus sparse matrices offers a powerful tool in solving computer vision problems. However, the results by the existing RPCA-based methods are often unsatisfied in complex scenes, for example varying illumination, shape changes, occlusions and shadows etc. In this talk, an adaptive RPCA which simultaneously preserves low-rank structure and restores the corrupted parts is proposed.  Specifically, the sum of weighted singular values is included in the objective function of minimization, and the weights are adaptively obtained by employing the proportion of information contained in corresponding singular values. Finally, we demonstrate several applications of the proposed methods in computer vision, such as image inpainting, shadow removal, background modeling and so on.