PINAR DUYGULU
Department of Computer
Engineering
Bilkent University
Ankara, Turkey, 06800
Phone: +90-312-290 31 43,
Fax:+90-312-266 40 47
e-mail :
duygulu[at]cs.bilkent.edu.tr
web:
http://www.cs.bilkent.edu.tr/~duygulu
Research Interests
Multimedia
Data Mining,
Computer Vision, Pattern Recognition, Statistical Machine Learning,
Information
Retrieval, Visual Perception.
Specifically,
large scale
object and face recognition, semantic analysis and retrieval of
largemultimedia collections, historical document analysis,
action recognition
Education
Ph.D.
(1998-2003)
Middle East Technical University,
Dep. of Computer Engineering, Ankara, Turkey
- Thesis title: Translating
images to words : A novel approach for object
recognition
- Supervisor: Prof.
Fatos Yarman-Vural
- Co-supervisor: Prof.
David Forsyth (U.C. Berkeley)
M.S.
(1996-1998)
Middle East Technical University,
Dep. of Computer Engineering, Ankara, Turkey
- Thesis title: Form
Document Identification Based On Hierarchical Tree Representation
- Supervisor: Assoc.
Prof. Volkan Atalay
B.S. (1992-1996)
Middle East
Technical University, Dep. of Computer Engineering, Ankara, Turkey
- Graduation Project: Recognition
of Cadastre Forms Using Form Description Languages
- Supervisors: Assoc.
Prof. Volkan Atalay and Asst. Prof. Halit Oguztuzun
Research Experience
Bilkent
University, Department of Computer Engineering
Assistant Professor (April 2004 - present)
- Group leader of the Turkish team for TRECVID2004, video
retrieval and analysis evaluation by NIST
- Team leader for "Basin Yayin Is Paketi" for joint YUUP
(Yayginlastirilmis Ulusal ve Uluslararasi
Proje) Project E-Devlet
- Researcher in MUSCLE - Multimedia
Understanding Through Semantics, Computation and Learning, a European
Commission FP6 Network of Excellence project
- Senior Researcher in Johns Hopkins University workshop series
on Joint Visual and Text Models, July-August 2004
- Working on the problem of associating different types of media,
currently audio transcripts and visual features, to solve the
correspondence problems between video frames and the related
transcripts. For example anchorperson/reporter talks about a person by
giving his/her name, but when the video related to that person is shown
usually the text does not include the name of the person.
- Working on the problem of finding the threads (evolution of a
story over time) and perspectives (different views of the same story)
on the news. Video clips of stories tend to be used over and over in
broadcast news. Tracking these duplicates (or similar videos) can be
used for understanding the evolution of a story in time or for
revealing different perspectives (for example Middle Eastern or
American) for the coverage of the story.
- Also, working on naming people (face recognition for a large
set) by integrating the face detection and audial/textual data.
Broadcast news are mostly based on people. However, current face
recognition system do not work very properly, and needs training for
each person. The aim is to learn face-name relationships from the
available data to find all the people in the news and to propose a
novel method for face recognition.
Johns Hopkins University, Center
for Language and Speech
Senior Researcher (July
2004 - August 2004)
- Worked in Joint Visual Text Modeling Team
Carnegie Mellon
University, Computer Science Department
Postdoctoral Researcher -
Informedia Project (March 2003 - March 2004)
- Worked for the Informedia Digital Video Understanding Project
which integrates speech, image and natural language understanding to
automatically transcribe, segment and index video for intelligent
search and image retrieval.The current Informedia library consists of
terabytes of data (broadcast news captured over the last years,
documentaries produced for public television and government agencies,
classroom lectures, and other video genres) with automatically
extracted metadata and indices.
- Involved in NIST TRECVID 2003 competition which aims
benchmarking content-based video retrieval systems. Specifically built
building and plane classifiers.
- Involved in writing proposals for VACE (video analysis and
content extraction for defense intelligence), AQUAINT(question
answering from errorful multimedia streams) and LifeLog(capturing and
analyzing personal experiences) projects.
- Developed a commercial detector for broadcast news videos and
integrated it into the project.
- Developed strategies to retrieve and browse the news stories
using both textual and visual features.
University
of California at Berkeley, Computer Science Division
Visiting Scholar - Supervisor :
Prof. David Forsyth (February 2001 - May 2002)
- Member of Computer Vision Group and Digital Library Project.
- Worked on linking image regions with associated text in the
large annotated data sets: namely Corel data set in which there are
40,000 images with annotated keywords and FAMSF data set in which there
are 80,000 images with associated text descriptions.
- Proposed a novel approach to object recognition on the very
large scale using the joint statistics of text and image regions.
Statistical machine translation ideas are applied to multimedia data
sets to translate image regions to words.
- Linked text and images for two novel applications:
auto-annotation (predicting words for the given images) which is very
helpful for retrieval of images and auto-illustration (suggesting
images for a given text passage).
- Built a system for browsing large image collections, more
specifically the large collection of images in Fine Art Museums of San
Francisco.
- Used Wordnet for disambiguating senses and finding the
hierarchy
of words.
Middle
East Technical University, Dep. of Computer Engineering
Research Assistant (September
1996 - March 2003)
- Member of the Image Processing and Pattern Recognition Group.
- Proposed multi-level segmentation and representation for
content based image retrieval. This method reduces the limitations of
under-segmentation and due to its hierarchical structure allows less
search space than oversegmentation.
- Worked on form document processing and retrieval. The
horizontal and vertical lines are used to represent the forms in a
hierarchical structure. Matching and retrieval of form documents are
proposed based on the hierarchical representations.
University
of Rochester
Visiting Researcher (July 1999)
Joint NSF-TUBITAK project with
Prof. Murat Tekalp and Prof. Fatos¸ Yarman-Vural.
- Worked on object based image retrieval. Multi-level
representation is used to search objects with parts in a hierarchical
manner.
Publications
A current list of publications can be found at
http://www.cs.bilkent.edu.tr/~duygulu/publications.html
Sponsored Projects
- Multimedia
data mining : Integration of multi-modal data for a better retrieval
and analysis
- TUBITAK Career Project, 04/2005-04/2010
- Semantic Multimodal Analysis of Digital
Media
- Sponsored by TUBITAK and European
Commission COST 292 Action, 10/2004-10/2008
- MUSCLE -
Multimedia
Understanding Through Semantics, Computation and Learning
- European
Commission FP6 Network of Excellence project, 2004-2008
- sponsored by Internal
Fellowship Programme for Creation of a large-scale image ontology
- IST-TURKEY: New
Information Society Technologies for Turkey (BTT-Turkiye: Turkiye Icin
Yeni Bilgi Toplumu Teknolojileri)
- Sponsored by TUBITAK,
12/2005-12/2008
- E-Devlet
- Joint YUUP Project
(Yayginlastirilmis Ulusal ve
Uluslararasi
Proje) with Middle East Technical University, Kocaeli
University, Suleyman Demirel University, Zonguldak
Karaelmas University, Anadolu University, Konya Selcuk University
- Sponsored by Turkish State
Planning Organization (Devlet Planlama
Teskilati - DPT)), 2004-2005
Teaching experience
Bilkent
University, Dept of Computer Engineering (2004-present)
- Courses given:
- CS 554 Computer Vision, Fall 2006, Spring 2006, Fall 2004,
Spring 2004
- CS 461 Artificial Intelligence, Spring 2007
- CS 315 Programming Languages, Fall 2004, Fall 2006
Middle East Technical University,
Dep. of Computer Engineering: Teaching Assistant (1996-1999)
- Courses Assisted :
Data Structures, Programming Language Concepts, Artifi- cial
Intelligence, Image Processing , Lisp and Prolog , Introduction to
Computers and Pascal Programming
Supervised students
- Nazli Ikizler, PhD
- Muhammet Bastan, PhD
- Selen Pehlivan, PhD
- Derya Ozkan, MSc
- Esra Ataer, MSs
- Tolga Can, MSc
- Kardelen Hatun, MSc
Professional activities
- Program Comittee for
- International Conference on Computer
Vision (ICCV) 2005,2007
- IEEE International Conference on Computer
Vision and Pattern
Recognition (CVPR) 2004-2006
- European Conference on Computer Vision
(ECCV) 2006
- ACM International Conference on Image
and Video Retrieval (CIVR) 2007
- ACM SIGMM International
Workshop on Multimedia Information Retrieval (MIR) 2006, 2007
- International Workshop on Multimodal
Information Retrieval in conjunction
with IJCAI 2007
- International RIAO Conference, 2007
- International
Symposium on Computer and Information Sciences (ISCIS), 2007
- Reviewer for the following journals
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- IEEE Transactions on Multimedia
- Pattern Recognition Letters
- Signal Processing
- IEEE Transactions on Circuits and Systems for Video Technology
- Reviewer for the following conferences
- Reviewer for TUBITAK Research and TEYDEP Projects
Honors and Awards
- TUBITAK Career Award
- Best paper in Cognitive Vision award in European Conference on
ComputerVision (ECCV 2002).
- NATO-PC Fellowship (A2) : Scholarship covering 6 month living
expenses plus travel.
- Graduated with the top score from High School.
Other activities
- EC Summer School, Bayesian Signal Processing, Newton
Institute, Cambridge,UK , 19-31 July 1998,
- NATO-ASI on Learning Theory and Practice (LTP 2002) July 8-19
2002, K.U. Leuven Belgium
- Researcher (Part Time) in TUBITAK BILTEN FORMAR Group
(1996-1997)
- Programmer(Part Time) in TUBITAK BILTEN Multimedia Group
(September1995- September 1996)
- Summer Internship in TUBITAK BILTEN Multimedia Group (1995)
- Summer Internship TRT (Turkish Radio and Television) Computer
Center (1994)
Invited talks
- Microsoft Research, Cambridge, “Object Recognition as Machine
Translation: Learning a Lexicon for a Fixed Image Vocabulary”, October
7, 2002.
Biographical
- Born: January 23, 1974 in Ankara, Turkey
- Married to Erol Sahin, mother of Gunes Ardic
- Languages spoken: Turkish (native), English (fluent), French
(beginner).
- Social activities involved:
- Guide Leader in IOI’99 (11. International Olympiad in
Informatics).
- Anchor, reporter and dubber in TRT (Turkish Radio and
Television) in several programs and news.
- Member of the METU Contemporary Dance Group (1991-1993)
- Member of Group Yore Turkish Folk Dance Group (2001-2002)