|
Zhipeng Zhao
Ph.D.
Department of Computer Science
Rutgers University
110
Frelinghuysen Road
Piscataway, NJ 08854-8019
zhipeng
AT cs.rutgers.edu
Lab/Office: CoRE 246
Curriculum Vitae (PDF, Word)
|
|
Bio
|
|
I am a freshly graduated Ph.D. from the Department of Computer Science, at
Rutgers University in New Jersey. My dissertation title is
“Towards a local-global visual feature-based framework for
recognition” and my advisor is Prof. Ahmed Elgammal. My research
interests include computer vision, machine learning and data mining.
I received M.S. degree in Statistics from Rutgers University
in 2007 and M.S. degree in Computer Science
from Old Dominion University in 2000. I
also received B.S degree in Computer
Science from Tsinghua
University, China in
1997.
|
|
Research
|
|
I have broad interests in computer vision, machine learning and data
mining. In particular, my research has been focused on general object
detection, recognition and human action analysis. In Computational Biomedicine
Imaging and Modeling Center (CBIM), I have worked as research assistant under
the supervision of Prof. Elgammal on general framework for object recognition
and human motion analysis. The projects I have worked on are:
- Integrate both
spatial and temporal information with appearance features for human
activity recognition: I designed and built a system for activity
recognition which modeled the human motions with the distribution of
local motion features and their spatial-temporal arrangements from the
discriminative key frames. This work could achieve an average 94.07%
recognition rate over benchmark dataset of human facial expressions,
hand gestures and general activities such as walking, running, boxing
etc. This work was published in FG08, ICPR08 (oral presentation), BMVC08 (oral presentation).
- Apply both
statistical and combinatorial methods for selecting informative parts to
build statistical models for part-based object recognition: I
designed and built a general object recognition system featuring a two
stage method for selecting local image features which characterize the
target object class. The first stage uses a combinatorial optimization
formulation for clustering on a weighted multipartite graph. The
following stage is a statistical method for selecting discriminative
patches from the positive images. This work improved recognition rate by
around 5% over the other known methods on a benchmark dataset and was
published at Beyond Patches
workshop in conjunction with CVPR06 and the extended journal version
was published in International Journal of
Computer Mathematics 07.
- Entropy based
vocabulary selection for local visual feature model: Two entropy
based methods for selecting informative vocabulary were implemented in
the “bag-of-meaningful-words” model for human activity
recognition. Both methods demonstrated performance improvements over the
baseline algorithm.
- Salient region
detection for video sequence via spectrum analysis: Applied energy
redistribution algorithm (e.g. logarithmic transformation) to the
amplitude spectrum of the video sequence for saliency detection. This
work is a generalization of previous published methods and could
increase the recognition speed.
|
|
Internship
|
|
I have worked as intern under the direction of Dr. Sandeep M Uttamchandani
in IBM Almaden research center on Rx project, an intelligent storage
management system:
·
User query recommendation tool using data
mining technology: Worked with a team of researchers in
developing and implementing a Problem Ticket Natural Language framework which
pre-processed the client’s problems with regards to IT system
management. Framework supported advanced features for learning-based
auto-complete and corrective query recommendations.
·
Fuzzy signature matching based on system
entity dependency: Implemented a fuzzy signature matching algorithm
which detected the system error signature from the diagnostic log.
Fuzzy logic was applied in the signature matching algorithm which supported
temporal and dependency constraints from the system dependency graph.
|
|
Selected Publications
|
- "Combinatorial
and Statistical Methods for Part Selection for Object Recognition ".
Zhipeng Zhao Akshay Vashist, Ahmed Elgammal, Ilya Muchnik and Casimir
Kulikowski . International Journal of Computer Mathematics,
volume 84 Issue 9, Sept. 2007.
Paper: [PDF-(745KB)]
- "Human
Activity Recognition from Frames Spatiotemporal Representation ".
Zhipeng Zhao and Ahmed Elgammal. International Conference on Pattern
Recognition (ICPR08), Dec. 2008. (oral presentation)
Paper: [PDF-(179KB)]
- "Information
Theoretic Key Frame Selection for Action Recognition ". Zhipeng
Zhao and Ahmed Elgammal. British Machine Vision Conference (BMVC08),
Sept 2008. (oral presentation with 12.5% accept rate)
Paper: [PDF-(102KB)]
- "Spatiotemporal
Pyramid Representation for Recognition of Facial Expressions and Hand
Gestures ". Zhipeng Zhao and Ahmed Elgammal. International
Conference on Automatic Face and Gesture Recognition (FG08), Sept
2008.
Paper: [PDF-(435KB)]
- " A
statistically selected Part-Based Probabilistic Model for Object
Recognition ". Zhipeng Zhao and Ahmed Elgammal. International
Workshop on Intelligent Computing in Pattern Analysis/Synthesis,
(IWICPAS06). , Xian, China,
August, 2006. LNCS 4153 pp 95-104.
Paper: [PDF-(543KB)]
- "
Discriminative Part Selection using Combinatorial and Statistical Models
for Part-Based Object Recognition ". Akshay Vashist , Zhipeng
Zhao, Ahmed Elgammal, Ilya Muchnik and Casimir Kulikowski Beyond
Patches Workshop in conjunction with CVPR06, vJune 2006.
Paper: [PDF-(967KB)]
|
|
Teaching
|
|
Teaching Assistent
|
|
|
cs510: Numerical Analysis
|
Spring 2008
|
|
|
cs510: Numerical Analysis
|
Fall 2007
|
|
|
cs534: Computer Vision
|
Spring 2005
|
|
|
cs510: Numerical Analysis
|
Fall 2004
|
|
|
cs111: Introduction to Computer Science
|
Summer 2004
|
|
|
cs111: Introduction to Computer Science
|
Spring 2004
|
|
|
cs111: Introduction to Computer Science
|
Fall 2003
|
|
|