法国CNRS博士后研究员李玺博士将来我校访问，并将给三次学术报告。三次报告均在信电系二楼215会议室举行。报告时间，题目，简介，及李博士简历如下。望有兴趣者踊跃参加。
谢谢留意！
Robust appearance modeling for visual tracking
March 27, 2009, Friday, 10am
Object tracking is a fundamental task of computer vision. It has many applications such as video surveillance, human computer interaction, human activity analysis, event detection, motion-based video indexing etc. In recent years, it becomes more and more demanding for public places such as banks, airports, plazas etc. Therefore, it is critically necessary and important to develop some robust systems for object tracking. However, some factors have hindered the development of object tracking in recent years, including self/inter-occlusions, pose changes, noises, varying lighting conditions and so forth. Consequently, it is very necessary for us to develop some robust appearance models which are immune to the aforementioned factors. Motivated by this, I have proposed four tracking methods for robust object appearance modeling. Theoretic analysis and experimental studies have demonstrated the effectiveness and robustness of our tracking methods under complex conditions.
Trajectory-based video retrieval using Dirichlet process mixture models
April 1, 2009, Wednesday, 3pm
Recent years have witnessed a drastic increase in motion based video retrieval applications. Among these applications, effective trajectory learning is a key issue to solve. Motivated by this, we present a trajectory-based video retrieval framework using Dirichlet process mixture models. The main contribution of this framework is four-fold. (1) We apply a Dirichlet process mixture model (DPMM) to unsupervised trajectory learning. DPMM is a countably infinite mixture model with its components growing by itself. (2)We employ a time-sensitive Dirichlet process mixture model (tDPMM) to learn trajectories’ time-series characteristics. Furthermore, a novel likelihood estimation algorithm for tDPMM is proposed for the first time. (3)We develop a tDPMM-based probabilistic model matching scheme, which is empirically shown to be more error-tolerating and is able to deliver higher retrieval accuracy than the peer methods in the literature. (4) The framework has a nice scalability and adaptability in the sense that when new cluster data are presented, the framework automatically identifies the new cluster information without having to redo the training. Theoretic analysis and experimental evaluations against the state-of-the-art methods demonstrate the promise and effectiveness of the framework.
Spectral clustering and linear discriminant analysis
April 2, 2009, Thursday, 10am
Spectral clustering is a powerful tool for unsupervised data learning. Most existing spectral clustering algorithms directly utilize the pairwise similarity matrix of the data to perform graph partitioning. Consequently, they are incapable of fully capturing the intrinsic structural information of graphs. To address this problem, we propose a novel random walk diffusion similarity measure (RWDSM) for capturing the intrinsic structural information of graphs. The
RWDSM is composed of three key components—emission, absorbing, and transmission. It is proven that graph partitioning on the RWDSM matrix performs better than on the pairwise similarity matrix of the data. Moreover, we present a novel discriminant graph partitioning criterion (DGPC) for fully capturing the discriminant information of graphs. The DGPC is designed to effectively characterize the intraclass compactness and the inter-class separability. Based on the RWDSM and DGPC, we further develop a novel spectral clustering algorithm (referred to as DGPCA). Theoretic analysis and experimental evaluations demonstrate the promise and effectiveness of the developed DGPCA.
Linear Discriminant Analysis (LDA) is a well-known scheme for supervised subspace learning. It has been widely used in applications of computer vision and pattern recognition. However, an intrinsic limitation of LDA is the sensitivity to the presence of outliers, due to using the Frobenius norm to measure the inter-class and intra-class distances. In this paper, we propose a novel rotational invariant L1 norm (i.e., R1 norm) based discriminant criterion referred to as DCL1), which better characterizes the intra-class compactness and the inter-class separability by using the rotational invariant L1 norm instead of the Frobenius norm. Based on the DCL1, three subspace learning algorithms (i.e., 1DL1, 2DL1, and TDL1) are developed for vector-based, matrix-based, and tensor-based representations of data, respectively. They are capable of reducing the influence of outliers substantially, resulting in a robust classification. Theoretic analysis and experimental evaluations demonstrate the promise and effectiveness of the proposed DCL1 and its algorithms.
Xi Li received the bachelor’s degree from Beijing University of Aeronautics and Astronautics with the major in electronic engineering in 2004. He then received the doctor’s degree from Institute of Automation, Chinese Academy of Sciences in January 2009. He is now a postdoctoral researcher of National Center for Scientific Research in France. He has published more than 20 international conference papers (including the top conferences such as ICCV, CVPR, ECCV), has developed several practical systems for commercial users, and has three national patents pending in review. Also, he has served as reviewers for many top conferences and leading journals. He is a creative person brave to take on new challenges.