· My main research interests are in two areas: (a) modeling and retrieval of digital video; and (b) multilingual text processing. I am concerned with the automatic indexing, retrieval and extraction of information from text and video The aim is to tackle difficult research issues in these areas to facilitate practical deployment of technologies to solve real-life problems.
· The long-term goal is to develop automated techniques to index an input video stream to facilitate the retrieval, summarization and personalization of video.
· To tackle these problems, we have developed a number of approaches:
a) A stratification model to represent the spatial-temporal video contents of video as layers of strata.
b) A temporal multi-resolution model for video segmentation that is able to handle all kinds of transition in a general framework.
c) A set of DCT-based techniques that uses the idea of contrast to extract and track objects, including faces and video text.
d) The use of cinematic rule, coupled with machine learning techniques, to locate story boundaries in video using a variety of features, including audio and text.
e) A learning-based approach to classify and segment story boundaries in news video using multiple features.
These will form the basis for video summarization and personalization.
· The long-term focus in this research is as follows:
a) To employ rigorous mathematical techniques to analyze and fuse multimedia contents for the purpose of classification, and high level feature extraction.
b) To investigate various learning-based and rule-induction techniques to segment news video into story units, and to detect events.
c) To develop adaptive techniques to personalize video based on users’ interests and browsing history for multimedia information retrieval on the web.
· The current projects are:
a) News video retrieval:
* To develop automated approach to index and retrieve news video.
* To participate in video TREC.
* Collaborators: I2R (Institute for Infocom Research) and Univ of Columbia.
b) VideoQA:
* This is an extension of news video project that aims to support interactive dialog with the users to retrieve personalized news video sequences.
* It integrates QA technology in text processing research with multimedia research.
· Publications in this area (click here).
· The focus is on developing techniques to segment, classify and extract information from free-text and web documents in both Chinese and English.
· We have built-up a platform for long-term research in information extraction. We have developed a system to perform event detection and information extraction -- for the web-based and free-text documents.
· We also started efforts on question-and-answer (QnA) research based on the platform we have built for information extraction. We have developed a system to perform TREC QA, and participated in TREC evaluation in August 2002.
· The long-term focus in text processing is as follows:
d) To research into the use of adaptive techniques to personalize text processing based on users’ interests and browsing history.
e) To develop effective techniques to support interactive self-service CRM, including techniques for dialogs, question-and-answer etc.
f) To investigate wrapper induction and other machine learning approaches to extract information on the web.
g) To research into cross language information extraction and question and answering.
h) To integrate interactive text processing techniques with video/audio.
· The current projects are:
a) Event-based QA
* To develop techniques to extract event elements in user queries to support both open and closed domain QA.
* To participate in QA-track in TREC.
b) Information Extraction on the web and free-text
* To extract or track specific entities or events on the web.
· Publications in this area (click here).
Objectives: to develop video retrieval system that uses content, spatial and temporal attributes for video representation and retrieval. A prototype news video retrieval system has been developed.
· Research grant funded by government agency
Principal Investigator of the ASTAR/MOE funded project on: Programme for Research into Intelligent Systems.
Approved fund: »2.9 million.
Duration: Jul 98 - Dec 02
Objectives: This is a large-scale project to research into techniques and framework for intelligent systems with applications in text and media processing areas. The project involves 3 academic staff, 4 Post-Doctorals, 6 Research Assistants, and over 10 graduate students.
(For further details, refer to http://www.comp.nus.edu.sg/~pris)
Project Outline: click here
| Back to Home | Back to Top |