Xu, Chao
Professor
Research Interests: Image processing
Office Phone: 86-10-6275 7094
Email: xuchao@cis.pku.edu.cn
Xu, Chao is a professor in the Department of Intelligence Science and technology, School of EECS. He obtained his B.Sc. from Tsinghua University in 1988, and Ph.D. from Institute of Electronics, Chinese Academy of Sciences in 1997 respectively. His research interests include video coding, image classification based on multi-view learning and multi-label learning, and image retrieval.
Dr. Xu has published more than 150 research papers, and many of them are published in top-tier conferences and journals, such as NIPS, CVPR, ICML, ACM MM, TPAMI, TIP, TKDE, and TNNLS. Among them, one paper is ESI Hot paper and five papers are ESI Highly Cited papers. He is serving as vice chair of intelligent information processing branch of China High-Tech Industrialization Association, and as a board member of digital signal processing branch of Chinese Institute of Electronics. He was awarded GE Education award (2000), Kodak Education award (2002), Zeng XianZi Education award (2016). His students, Yong Luo was awarded CCF Best Ph.D. Paper award (2016) and had 3 ESI Highly Cited papers, and Bo Geng was awarded Peking University Best Ph.D. Paper award (2012) and ACM CIKM Best Student Paper award, now is the CTO of mi-u company, and Chang Xu were awarded Peking University Best Ph.D. Paper award (2016) and had 1 ESI Hot paper and 1 Highly Cited papers, now is a lecturer in The University of Sydney.
Dr. Xu has more than ten research projects including NSFC, 973 programs, 863 project, etc. His research achievements are summarized as follows:
1) Video coding techniques: One major research topic in video processing field is to preserve higher image quality with less data, make the video management system support more video storage and faster capture and transportation. He proposed some new video coding techniques, including pipeline wavelet transformation, parallel bit-plane encoding, primitives grouping architecture, to design high speed video compression system (120f/s) for China Aerospace Corporation, and to accomplish the first Chinese digital cinema system for DADI Digital Cinema Corporation (now the second largest cinema company in China), and to support the video conference system for Datang Communication Corporation.
2) Image classification based on multi-view learning: Natural images are taken from different cases, such as view point, scale and illumination, etc. Multi-view learning combines various information to improve the performance of natural image classification. He focused on the feature space learning techniques for different image data, and proposed some new solutions, including intact feature space learning (TPAMI 2015, ESI Hot paper), background removal based on information bottleneck method (TPAMI 2014, ESI Highly Cited paper), ensemble manifold regularization (TPAMI 2012, ESI Highly Cited paper). These methods can not only improve image classification, but also reconstruct the object image, and alleviate background interference, or take semi-supervised learning.
3) Image classification based on multi-label learning: Natural images usually contain multiple objects, such as human and car, cloud and sky. Multi-label learning exploits the relationship between objects to improve the performance of image classification. He proposed new multi-label learning techniques, including manifold regularized multitask learning (TIP 2013, ESI Highly Cited paper), multi-view vector-valued manifold regularization for multi-label learning (TNNLS 2013, ESI Highly Cited paper), and multi-view matrix completion for multi-label learning (TIP 2015, ESI Highly Cited paper). These methods improve natural image classification with multitask learning and combination of multi-view learning and multi-label learning.
