Jingen Liu

Computer Vision Lab,
School of Electrical Engineering and Computer Science,
University of Central Florida,
Orlando, FL, USA
Email: liujg - at - cs - dot - ucf -dot - edu

I am a Ph.D student in Computer Vision Lab at University of Central Florida, under supervision of Dr. Mubrak Shah. My current research interests are scene understanding and recognition, action recognition, video content analysis and retrieval, object recognition, and crowd tracking. Before I joined the vision lab, I worked in coordination computing with Dr. Dan C. Marinescu from 2004 to 2005 .
I got my bachelor and master degree from Huazhong University of Science and Technology, Wuhan, China.

PUBLICATIONS

Kishore Reddy*, Jingen Liu* and Mubarak Shah, Incremental Action Recognition Using Feature Tree, to appear in IEEE International Conference on Computer Vision (ICCV), Kyoto, Japan, 2009[project page]

Yang Yang, Jingen Liu and Mubarak Shah, Video Scene Understanding with Multi-Scale Analysis, to appear in IEEE International Conference on Computer Vision (ICCV), Kyoto, Japan, 2009 [project page]

Jingen Liu, Yang Yang & Mubarak Shah, Learning Semantic Visual Vocabularies Using Diffusion Distance, to appear in IEEE International Conference on Computer Vision and Pattern Recognition(CVPR), Miami, 2009.

Jingen Liu, Jiebo Luo & Mubarak Shah, Recognizing Realistic Actions from Videos "in the Wild", to appear in IEEE International Conference on Computer Vision and Pattern Recognition(CVPR), Miami, 2009. [dataset]

Jingen Liu, Jiebo Luo & Mubarak Shah, Action Recognition in Unconstrained Amateur Videos, to appear in IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Taipei, 2009.

Fangshi Wang, Wei Lu, Jingen Liu & Mubarak Shah, Automatic Video Annotation with Adaptive Number of Key Words, IEEE International Conference on Pattern Recognition (ICPR), Tampa, 2008.

Jingen Liu & Mubarak Shah, Learning Human Actions via Information Maximization, IEEE International Conference on Computer Vision and Pattern Recognition(CVPR), Alaska, 2008.

Jingen Liu, Saad Ali & Mubarak Shah, Recognizing Human Actions Using Multiple Features, IEEE International Conference on Computer Vision and Pattern Recognition(CVPR), Alaska, 2008.

O. Bilal Orhan, Jingen Liu, et al., University of Central Florida at TRECVID 2008: Content Based Copy Detection and Surveillance Event Detection, TREC Video Retrieval Evaluation Workshop (TRECVID), Gaithersburg, November 2008.

Jingen Liu & Mubarak Shah, Scene Modeling Using Co-Clustering , IEEE International Conference on Computer Vision (ICCV), Rio de Janeiro, Brazil, 2007.

Yun Zhai, Jingen Liu & Mubarak Shah, Automatic Query Expansion in News Video Retrieval, IEEE International Conference on Multimedia and Expo (ICME), Toronto, Canada, 2006.

Andrew Miller, Aslan Basharat, B. White, Jingen Liu & Mubarak Shah, Person and Vehicle Tracking in Surveillance Video. CLEAR Evaluation Campaign and Workshop, Baltimore , 2007.

Jingen Liu , Yusuf Aytar, Bilal Orhan, Jenny Han & Mubarak Shah, University of Central Florida at TRECVID 2007: Semantic Video Classification and Automatic Search, TREC Video Retrieval Evaluation Workshop (TRECVID), Gaithersburg, 2007.

Jingen Liu , Yun Zhai & Mubarak Shah, PEGASUS: An Information Mining System for TV News Videos, SPIE Defense and Security Symposium, Orlando, 2006.

Jingen Liu , Yun Zhai, Arslan Basharat, Bilal Orhan, Saad Khan, Humera Noor, Phillip Berkowitz and Mubarak Shah, University of Central Florida at TRECVID 2006: High-Level Feature Extraction and Video Search, TREC Video Retrieval Evaluation Workshop (TRECVID), Gaithersburg, 2006.

Yun Zhai, Jingen Liu, Xiaochun Cao, Arslan Basharat, Asaad Hakeem, Saad Ali, Mubarak Shah, Costantino Grana & Rita Cucchiarra, Video Understanding and Content-Based Retrieval, TREC Video Retrieval Evaluation Workshop (TRECVID), Gaithersburg, Maryland, 2005.

[Previous]

Jingen Liu, et. al., The Index Parallelization Technique in Parallel Query System, Journal of Huazhong Univ. of Sci. & Tech. (Nature Science), 2003, Vol. 31, No. 1 (In Chinese) .

Jingen Liu, et. al., A Software Testing Technology Based On Class Data Flow, Journal of Huazhong Univ. of Sci. & Tech (Nature Science), 2003, Vol. 31 No. 11 (In Chinese).

Jianguo Dang, Weiwei Yang, Jingen Liu, Time-Sequence Handling in PVM Parallel debugger DPD, Journal of Huazhong Univ. of Sci. & Tech. (Nature Science), 2003, Vol. 31, No. 3 (In Chinese).

RESEARCH PROJECTS

	Learning Semantic Vocabularies using Diffusion Distance We use diffusion maps to automatically learn a semantic vocabulary from abundant quantized mid-level features. Each mid-level feature is represented by the vector of point-wise mutual information (PMI). In this mid-level feature space, we believe the features produced by similar sources must lie on a certain manifold. To capture the intrinsic geometric relations between features, we measure their dissimilarity using diffusion distance. We have applied this technique to action recognition, video scene understanding (activity discovery), crowd tracking.... [project page]

	CLEAR Evaluation Campaign , CHIL&NIST I worked as an active member of the UCF CLEAR evaluation team during 2007, on tasks as people and vehicle tracking in surveillance video. We applied our KNIGHT system to detect and track the moving objects. Classfiers are used to recognize the object. My major contribution in this evaluation is designing effective people and vehicle classifiers on the detected objects [report 2007].

	Learning Realistic Actions from Unconstrained Videos We developed a systematic framework for recognizing realistic actions from unconstrained videos which have tremendous variations due to camera motion, background clutter, changes in object appearance and scale, etc. We use both motion and static features from the videos. Since the raw features of both types are dense yet noisy, we propose strategies to prune these features...... [project page]

	Action Recognition Using Bag of Video-Words Developed an algorithm that used mutual information between the cuboids (3D interest points) and videos to automatically discover the optimized number of video-word-cluster while balancing the discrimination and compactness of the model. The human actions are modeled by the video-word-clusters. Our approach is able to handle camera motion, changes of illumination and appearance, and zooming etc. [project page]

	Action Recognition Using Multiple Features We extracted spatiotemporal features (cuboids) and Spin-Image features (which is generated from the Space-time volume) from the action object. A general framework based on Fiedler Embedding was developed to fuse the two complementary features for human action recognition.[project page]

	Scene Understanding and Classification Beyond bag of features, there exists some hidden relationship between the low level image patches. Those hidden relationships are called semantic intermediate concepts. We utilize the Maximization of Mutual Information co-clustering approach to automatically discover the intermediate concepts. The scene recognition is conducted by exploring the statistic information of the intermediate concepts [project page].

	TREC Video Retrieval Evaluation, NIST I worked as an active member of the UCF T RECVID team during 200 5 and team leader during TRECVID 200 6 and 2007. The tasks which I was involved in are high-level feature extraction and video retrieval . During 2006, our performance on high-level feature task is top 5 of 35 teams, and for interactive search our run ranks top 4 of 36 total runs from the community. During 2007, our performances on high level feature and automatic search task were above the median.[report 2005, report2006, report2007]

	Features Fusion through Fiedler Embedding Developed a principled framework for fusing information from complementary features (patches, contours, segments) for object recognition [project page].

	Image Classification Using Boosted Hidden Features We automatically extracted hidden features from multiple low level features including spatial color statistics, spatial edge histogram, and SIFT features, which contains both the local and global information of the image. In stead of assigning equal weights to the hidden features, AdaBoost creates the classifier by automatically select the prominent features for each category.

	PEGASUS: A Content-Based Video Retrieval System We created the PEGASUS which is an interactive video retrieval system. It is designed for retrieving relevant video shots from the broadcast news programs given the target topics. It contains three components: web-based user interface, server and fast feature index system. Both the speech transcript feature and visual features have been explored.System web http://pegasus.cs.ucf.edu:8080.[project page]

	Consumer Video Indexing, Kodak In this project we worked on developing a system for consumer video indexing. We explored two approaches for two sorts of videos, e.g. physical scene videos and action videos. For scene videos, we detected video shot boundaries and indexed the shots by the key frames of the shot. For action videos, we explored the kinematic features of the optical flows of the action video, and videos are indexed by these invariant flow features

	Coordination Computing (under supervision of Dr. Marinescu) We worked on developing a coordination system based on Condensed Graph and Petri Net for distributed task scheduling. We set up the “UCF Summer Grid” as our experiment platform using the NFS Initiative Middleware .