About Me

Experience

Research

Awards

Publications

Activities

Links

Yue GAO

Lab for Media Search
School of Computing
National University of Singapore

Room 525, AS6
National University of Singapore
13 Computing Drive
Singapore 117417

Emails:  kevin.gaoy@gmail.com
gaoyue@comp.nus.edu.sg

About Me

I am currently a research intern in Lab for Media Search, School of Computing, National University of Singapore, and my advisor is Tat-Seng Chua. I am working on tag and geotag based applications, and 3D model retrieval.

My CV is available now. (pdf)

Education

    M.Eng.                                                                                Sep. 2005 - Jun. 2008
    Tsinghua University                                                                        Beijing, China 

    B.Eng.                                                                                Sep. 2001 - Jun. 2005
    Harbin Institute of Technology                                           Harbin, China

Research Interest

• Computer Vision,        including scene classification and 3D object analysis/retrieval
  
• Multimedia Content Analysis,        including video summarization and image/video retrieval
  
• Social Media,        including recommendation system and tag analysis
  

Research

Social Media

LocGo: A dynamic generation of travel guide by automatic landmark ranking Work with Tat-Seng Chua, Ramesh Jain, and Jinhui Tang,Richang Hong. In this work, we present a travel guide system LocGo (Location Go), which can automatically recognize and rank the landmarks for travelers. In this system, a novel Automatic Landmark Ranking (ALR) method is proposed by utilizing the tag and geo-tag information of photos in Flickr and user knowledge from Yahoo Travel Guide. The goal of the proposed LocGo is to provide a ranked landmark list and corresponding representative views to users for requested location name. Related paper: Yue Gao, Jinhui Tang, Richang Hong, Tat-Seng Chua, Ramesh Jain. ''LocGo: A Dynamic Generation of Travel Guide by Automatic Landmark Ranking'', submitted to WWW2010.

Video Segmentation, Representation, and Retrieval

Video shot representation and application for shot retrieval Work with Jinhui Tang, and Xudong Xie. Addressing the problem of video representation, a key frame vector (KFV) is designed, which considers both temporal information and video content weight of shots. Temporal correlation between frames reflects the relation between visual content of frames. Video retrieval needs comparison between shots. KFV employs a block-based frame comparison scheme, and when compare two shots, it uses EMD to compare shots represented using KFVs. Related paper: Yue Gao, Jinhui Tang, Xudong Xie. ''Key Frame Vector and Its Application Shot Retrieval'', ACM International Workshop on Interactive Multimedia for Consumer Electronics (IMCE2009, In association with ACM Multimedia 2009)

Video shot detection and video scene segmentation Video segmentation is the basic of content based video analysis. To build a video analysis system, we have implemented a shot detection algorithm using information theory combining several recent shot detection works. Technical Reports: Yue Gao''Video Parsing: Shot Detection and Scene Segmentation'', School of Software, Tsinghua University, 2006.

3D Object/Model Retrieval

	3D model comparison using spatial structure circular descriptor The spatial structure is important in content-based 3D model analysis. We design a new view-based 3D model retrieval algorithm. Within the SSCD,the spatial structure of a 3D model is described by 2D images, and the attribute values of each pixel represent 3D spatial information. Hence, SSCD can preserve the global spatial structure of 3D models, and is invariant to rotation and scaling. In addition, by using 2D images to describe the spatial information of 3D models, all spatial information of the 3D models can be represented by SSCD without redundancy. Thus, SSCD can be applied to many scenarios which utilize spatial information. Also, we design an SSCD-based 3D model comparison algorithm. Related paper: Yue Gao, et al., ''3D Model Retrieval using Spatial Structure Circular Descriptor'', Pattern Recognition, to appear.
	View-based 3D model retrieval using probabilistic graph method In this work, we design a novel view-based 3D model retrieval algorithm. Five circle camera arrays are employed in this work, and five groups of views are captured for each 3D model. We model each view group as a first order Markov Chain, and the task of 3D model retrieval is defined as a probabilistic analysis procedure. The comparison between the query and other 3D models is changed to compute the conditional probabilities of 3D models in the database given the query model. Related paper: Yue Gao, Jinhui Tang, Haojie Li, Qionghai Dai, Naiyao Zhang. ''View-based 3D Model Retrieval with Probabilistic Graph Model'', Neurocomputing, to appear.
	View-based 3D model retrieval with different camera array setting Work with Tat-Seng Chua, and Jinhui Tang. Traditional view-based 3D model retrieval algorithms depend on a predefined camera arrays. How to cope with the situation when the camera arrays are different or not specified? In this work, we propose a probabilistic framework to retrieve 3D obejcts using free view sets. This is the first attempt towards view-based 3D object retrieval cross different camera arrays. In this work, Zernike moment is employed, and the feature distribution is modeled using gaussian models. First all existing views of the query object are grouped to clusters. Then the feature distribution is trained in each cluster. The probabilistic framework is employed to compute the posterior probability of each object given the query. Related paper: A draft version is available.
	Image-based 3D object retrieval using free view sets Work with Tat-Seng Chua, and Jinhui Tang. In real life, it is difficult to capture all 3D information (or all views) foe every 3D object. Therefore, traditional view-based method cannot be applied directly into real applications. In this work, we propose a probabilistic framework to retrieve 3D obejcts using free view sets. To our knowledge, this is the first attempt towards view-based 3D object retrieval without full model information.

Video Summarization

	Clip-based video summarization and ranking In this work, we design a proportional maxweighted bipartite matching algorithm, and apply it to video clip matching. This method first generates a basic frame set and a corresponding proportion value set from each video clip. Then it models two clips as a weighted bipartite graph, where the weight values are determined by both the direct frame similarities and the proportion values. Then the max-weighted bipartite matching is employed to measure the similarity between two clips. This method achieves good retrieval performance when the length of two clips varies greatly. We further apply it to video clip summarization. Related paper: Yue Gao, et al., ''Clip-based Video Summarization and Ranking'', ACM International Conference on Image and Video Retrieval (CIVR2008), Pages 135 - 140, 2008.
	Home video summarization In this work, we design a video summarization algorithm using redundancy detection and content ranking. The redundancy removing is based on key frame clustering and repetitive segment detection. The time-constraint summary is constructed using the important factor criterion. Related paper: Tao Wang, Yue Gao, et al., ''Video Summarization by Redundancy Removing and Content Ranking'', ACM Conference on Multimedia (MM2007),Pages 577 - 580, 2007.
	Rush summarization In this work, we design five detectors to remove useless and low-quality frames. Color bars, Near-monochrome frames, the abrupt frames, the shaking frames and the clap board frames can be detected effectively. We have participated the TRECVID competition on Rush Summarization 2007, and we achieved a good performance on shortening the video length. Related paper: Tao Wang, Yue Gao, et al., ''THU-ICRC at Rush Summarization of TRECVID 2007'', the international workshop on TRECVID video summarization (In association with ACM Multimedia 2007), Pages 79 - 83, 2007.

Awards

Publications

                                          Excellent Master Thesis Award in Tsinghua University

Activity

LINKS

 Statistics since 2009.09.29

 Visits since Sep. 28, 2009  Visitors since Sep. 28, 2009