Page 56 - profile2014.indd

特聘研究員
廖弘源 Hong-Yuan Mark Liao

Distinguished Research Fellow
Ph.D., Electrical Engineering, Northwestern University
Tel: +886-2-2788- 3799 ext.1550 or 1519 Fax: +886-2-2789-3309
Email: liao@iis.sinica.edu.tw
http://www.iis.sinica.edu.tw/pages/liao/index_en.html

● Multimedia Chair Professor, Research Description
National Chung Hsing Univ.
(2009-12)
Over the past 20 years, my research has focused on multimedia signal processing and mul-
● Adjunct Chair Professor, College timedia security. I will describe here my major achievements in the eld of multimedia up

of ECE, National I-Lan Univ. to the time of writing:
(2010-11, 2013-14)
● Adjunct Chair Professor, Chung Face Recognition
Yuan Christian Univ. (2010 - )
During 1997-2001, I led a project on face detection and recognition. My former Ph.D. stu-
● IEEE Fellow (2013) dent (Prof. Li-Fen Chen of National Yang-Ming University) and I initially developed a Linear
● Academia Sinica Young Investi- Discriminant Analysis (LDA)-based face recognition system that can solve the small sample
gator Award, Academia Sinica size problem. We fully utilized the concept of linear algebra to prove that the most expres-
(1998) sive vectors derived in the null space of the within-class scatter matrix using principal com-
● Distinguished Research Award, ponent analysis (PCA) are equal to the optimal discriminant vectors derived in the original
National Science Council (2003- space using LDA. This work later became a milestone paper, and is referred to as the null-
06, 2010-13, 2013-16) space LDA by numerous subsequent studies. According to Thomson Reuters’ ESI report, this
● Academia Sinica Investigator paper has been a top 0.1% highly cited paper in the Computer Science Category (ISI). In
Award, Academia Sinica (2010-14) 2000, we solved an important problem related to face database construction. We proposed

● Associate Editor, IEEE Trans. on a statistics-based technique to quantitatively prove that the in uence of the non-face por-
Multimedia (1998 – 2001), Image tion could be much larger than the “real’’ face portion, causing incorrect results. This work
Processing (2009-13), Information has been selected by HYPERLINK “http://www.face-rec.org/”www.face-rec.org as one of the
Forensics and Security (2009-12)
most interesting papers published in the past two decades.
● Editorial Board Member, IEEE
Signal Processing Magazine Digital Watermarking
(2010-13), EURASIP Journal on During 1998-2002, my former postdoctoral fellow, Dr. Chun-Shien Lu, and I proposed ways
Advances in Signal Processing
(2006- ), Journal of Visual Com- to design robust and/or fragile digital watermarking methodologies. One such methodol-
munication and Image Represen- ogy, the cocktail watermarking technique, is a seminal method which addressed the major
tation (2000-10) aws in two types of conventional methods – those based on spread spectrum and others

● Division Chair, Computer Science based on human visual models. This technique has been well recognized worldwide. The
(2), National Science Council theoretical part of this work was published in IEEE Transactions on Multimedia 2000 and it
(2008-2011) has been a top 1% highly cited paper (ISI). In a second project, we proposed a novel multi-
● Region 10 Director (Asia-Pacific), purpose watermarking scheme, in which robust and fragile watermarks are simultaneously
IEEE Signal Processing Society embedded, for copyright protection and content authentication. This work was published
(2014-2015)
in IEEE Transactions on Image Processing 2001 and it has been a top 1% highly cited paper.
● BS, Physics, National Tsing Hua
University (1981), MS (1985), Video Surveillance
Ph.D. (1990), EE, Northwestern During 2004-2008, my team worked on video surveillance. We proposed video surveillance
University

techniques based on human movement analysis. In our rst system, each sequence of hu-
man movement is converted into a posture sequence. For each posture, we extracted two
features: the skeleton feature and the centroid context feature. The two descriptors work-
ing together make human movement analysis a very e cient and accurate process. This

work was published in IEEE Transactions on Multimedia (Vol. 10, No. 3, April 2008). Sub-
sequently, my team further advanced the technique of human movement analysis to the
atomic action level. This work was published in IEEE Transactions on Systems, Man, and
Cybernetics, Part B, in 2009. The main accomplishment of this work is that we successfully

made use of the variable length Markov model (VLMM) to handle motions with di erent
speeds. This problem could not be handled by the conventional hidden Markov model.

56 特聘講座/特聘研究員 Distinguished Chair and Distinguished Research Fellows