|
|
Current Projects
High-Level Feature Extraction for TRECVID Benchmark
Introduction
The high-level feature extraction task aims at detecting semantic concepts from video
shots, including objects like car, building, scenes like waterscape-waterfront, events
like parade & demonstration, etc. For each video shot, a probability hypothesis is
given for each concept indicating the chance of concept occurrence. Accordingly for
each concept, video shots are ranked based on these probabilities in descending order.
Precision values at different recall positions are calculated for evaluation.
Since 2006, I have been participating in the TRECVID high-level feature extraction task.
- For TRECVID 2006, two pieces of my works were incorporated into the Columbia's high-level
feature extraction submission. (1) context-based concept fusion by incorporating inter-conceptual
relationships by a Conditional Random Field; (2) construct Lexicon-Spatial Pyramid Match (LSPM)
kernels with local SIFT descriptors for SVM based classification.
- For TRECVID 2007, we studied and evaluated different cross-domain learning algorithms in the context
of semantic concept detection over a large-scale video set. Our approach -- the Cross-Domain SVM (CDSVM)
algorithm was incorporated into the Columbia's high-level feature extraction submission.
- For TRECVID 2008, my algorithm -- Principle Component Semi-Supervised SVM (PCSVM) was incorporated to
the IBM's high-level feature extraction submission, where the algorithm tries to jointly optimize feature
subspace learning and SVM learning.
Publications
- Apostol Natsev, Wei Jiang, Michele Merler, John Smith, Jelena Tesic, Lexing Xie, Rong Yan,
"IBM research TRECVID-2008 video retrieval system", NIST TRECVID Workshop, 2008. PDF
- Shih-Fu Chang, Wei Jiang, Akira Yanagawa, Eric Zavesky, "Columbia university TRECVID
2007 high-level feature extraction", NIST TRECVID Workshop, 2007.
PDF
- Shih-Fu Chang, Winston Hsu, Wei Jiang, Lyndon Kennedy, Dong Xu, Akira Yanagawa, Eric Zavesky,
"olumbia university TRECVID-2006 video search and high-level feature extraction", NIST TRECVID Workshop, 2006.
PDF
|
