Page 54 - 2017 Brochure
P. 54
聘研究員
Distinguished Research Fellows
廖弘源 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/
Research Description novel multipurpose watermarking scheme for copyright
protection and content authentication, where robust and
For the past 25 years, my research has focused on fragile watermarks are simultaneously embedded. This work
multimedia signal processing and multimedia security. was published in IEEE Transactions on Image Processing in
Herein, I report our major achievements in these fields: 2001 and it has been a top 1% highly cited paper.
Facial Recognition Video Surveillance
From 1997 to 2001, I led a project on face detection In the years between 2004 and 2008, my team worked
and recognition. The first contribution that my former on video surveillance. We proposed methods for video
Ph.D. student (Prof. Li-Fen Chen of National Yang-Ming surveillance based on human movement analysis. In the first
University) and I made was a Linear Discriminant Analysis system we developed, each sequence of human movement
(LDA) based face recognition system that can solve the is converted into a posture sequence. For each posture, we
small sample size problem. We utilized concepts within extract two features, the skeleton and the centroid context.
linear algebra to prove that the most expressive vectors These two descriptors, working together, make human
derived in the null space of the within-class scatter matrix movement analysis a very efficient and accurate process.
from principal component analysis (PCA) are equal to the This work was published in IEEE Transactions on Multimedia
optimal discriminant vectors derived in the original space in 2008. Subsequently, my team further advanced the
from LDA. This work became a milestone paper and is technique of human movement analysis to the atomic
referred to as the ‘null-space LDA’ by numerous subsequent action level. This work was published in IEEE Transactions
studies. According to Thomson Reuters ESI, this paper is on Systems, Man, and Cybernetics, Part B, in 2009. The
among the top 0.1% of highly cited papers in the category main contribution of this work was that we made use of
of Computer Science. We also solved an important problem the variable length Markov model to handle motions with
related to face database construction. We proposed a different speeds. This problem could not be resolved using
statistics-based technique to quantitatively prove that the the conventional hidden Markov model.
influence of the non-face portion could be much larger than
the ‘real’ face portion, causing incorrect results. This work Video Inpainting
has been selected by www.face-rec.org as one of the most My group started working on this topic in 2009, when
interesting papers published in the past two decades. executing the Taiwan e-Learning and Digital Archives
Project (2008-2012). In recent years, recording cultural and
Digital Watermarking historical artifacts, such as photographs and films or videos,
During the period between 1998 and 2002, my former in digital formats has become an important undertaking.
postdoctoral fellow, Dr. Chun-Shien Lu, and I proposed However, because of their age, the visual quality of such
ways to design robust and/or fragile digital watermarking images and videos after digitization is usually very poor,
methodologies. One such methodology, the cocktail often exhibiting unstable luminance and damaged content.
watermarking technique, became a seminal method which Video inpainting, an extremely challenging technique,
addressed the major flaws in two types of conventional helps users remove undesirable objects and repair areas
methods that are either based on spread spectrum or where content is missing or damaged. We have contributed
human visual models. This technique has gained worldwide several important works to this field. First, since human
recognition. The theoretical part of this work was published motion is articulated by nature, we proposed to solve this
in IEEE Transactions on Multimedia in 2000 and is among
the top 1% of highly cited papers. We also proposed a
52 特聘講座 / 特聘研究群 Distinguished Chair and Distinguished Research Fellows
Distinguished Research Fellows
廖弘源 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/
Research Description novel multipurpose watermarking scheme for copyright
protection and content authentication, where robust and
For the past 25 years, my research has focused on fragile watermarks are simultaneously embedded. This work
multimedia signal processing and multimedia security. was published in IEEE Transactions on Image Processing in
Herein, I report our major achievements in these fields: 2001 and it has been a top 1% highly cited paper.
Facial Recognition Video Surveillance
From 1997 to 2001, I led a project on face detection In the years between 2004 and 2008, my team worked
and recognition. The first contribution that my former on video surveillance. We proposed methods for video
Ph.D. student (Prof. Li-Fen Chen of National Yang-Ming surveillance based on human movement analysis. In the first
University) and I made was a Linear Discriminant Analysis system we developed, each sequence of human movement
(LDA) based face recognition system that can solve the is converted into a posture sequence. For each posture, we
small sample size problem. We utilized concepts within extract two features, the skeleton and the centroid context.
linear algebra to prove that the most expressive vectors These two descriptors, working together, make human
derived in the null space of the within-class scatter matrix movement analysis a very efficient and accurate process.
from principal component analysis (PCA) are equal to the This work was published in IEEE Transactions on Multimedia
optimal discriminant vectors derived in the original space in 2008. Subsequently, my team further advanced the
from LDA. This work became a milestone paper and is technique of human movement analysis to the atomic
referred to as the ‘null-space LDA’ by numerous subsequent action level. This work was published in IEEE Transactions
studies. According to Thomson Reuters ESI, this paper is on Systems, Man, and Cybernetics, Part B, in 2009. The
among the top 0.1% of highly cited papers in the category main contribution of this work was that we made use of
of Computer Science. We also solved an important problem the variable length Markov model to handle motions with
related to face database construction. We proposed a different speeds. This problem could not be resolved using
statistics-based technique to quantitatively prove that the the conventional hidden Markov model.
influence of the non-face portion could be much larger than
the ‘real’ face portion, causing incorrect results. This work Video Inpainting
has been selected by www.face-rec.org as one of the most My group started working on this topic in 2009, when
interesting papers published in the past two decades. executing the Taiwan e-Learning and Digital Archives
Project (2008-2012). In recent years, recording cultural and
Digital Watermarking historical artifacts, such as photographs and films or videos,
During the period between 1998 and 2002, my former in digital formats has become an important undertaking.
postdoctoral fellow, Dr. Chun-Shien Lu, and I proposed However, because of their age, the visual quality of such
ways to design robust and/or fragile digital watermarking images and videos after digitization is usually very poor,
methodologies. One such methodology, the cocktail often exhibiting unstable luminance and damaged content.
watermarking technique, became a seminal method which Video inpainting, an extremely challenging technique,
addressed the major flaws in two types of conventional helps users remove undesirable objects and repair areas
methods that are either based on spread spectrum or where content is missing or damaged. We have contributed
human visual models. This technique has gained worldwide several important works to this field. First, since human
recognition. The theoretical part of this work was published motion is articulated by nature, we proposed to solve this
in IEEE Transactions on Multimedia in 2000 and is among
the top 1% of highly cited papers. We also proposed a
52 特聘講座 / 特聘研究群 Distinguished Chair and Distinguished Research Fellows
