Current Research:

To see the details of all these topics, please click here (it takes time to load all the images).

Automatic Body Parts Segmentation and Surveillance Video Retrieval:

Given a surveillance video with pedestrian walking along similar paths, body part segmentation is obtained automatically using geometric transform (GeT) proposed in TIP GET.The color features of each part are then extracted and used to retrieve videos of pedestrians with similar wears. Details.

　

　

　

　

Fingerprinting human and vehicles in videos using geometric transform:

We use the geometric transform (GeT) based appearance modeling to extract the signatures of human and vehicles in the video. GeT is used to incorporate geometric prior information into appearance modeling, so that we can deal with variations in poses and views. This method can be extended to track objects persistently through multiple cameras. (ICCV 05 TIP GET)

The key in matching pedestrian's appearance is to normalize the appearance of pedestrians despite different poses and shapes of each subject. Geometric transform (GeT) based on shape matcing provides a way to synthesize 'normalized' appearance at arbitrary poses. Its advantage is similar to Active Appearance Model (AAM), which removes the effect of shapes when comparing appearances. But here we synthesize the 'normalized' appearances of objects with articulated motion, a case hard for AAM. Details.

　

Structure from planar motion (SfPM):

Planar motion is ubiquitous in surveillance videos, because objects are usually moving on a ground plane. So we aim at developing a method specialized for 3D reconstruction from planar motions. A fully automated system for 3D vehicle model reconstruction is built. The extracted model can be used for vehicle identification, computer graphics applications, or other activity analysis. ( MOTION 05(Oral), TIP 2006) Details.

　

　

　

　

Level set segmentation:

We implemented the level set segmentation based on texture and motion. Details.

　

　

　

Previous Research Topics:

Face recognition in compressed imagery: 

The effects of image and video compression on face recognition in the still-to-video setting are studied. We use particle filter to tracking and recogzing face simultaneously. To deal with the illumination and pose variations in the outdoor sequences, intrapersonal space (IPS) is constructed from examplar views and used to calculate the likelihood density. Both the gallery images and probe videos are compressed and several experiments are run to study their effects on the recognition rate. (ICASSP 04)

A comparison of subspace analysis for face recogntion:

We report the results of a comparative study on subspace analysis methods for face recognition. In particular, we have studied four different subspace representations and their `kernelized' versions if available. They include both unsupervised methods such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA), and supervised methods such as Fisher Discriminant Analysis (FDA) and probabilistic PCA (PPCA) used in a discriminative manner. The `kernelized' versions of these methods provide subspaces of high-dimensional feature spaces induced by non-linear mappings. To test the effectiveness of these subspace representations, we experiment on two databases with three typical variations of face images, i.e, pose, illumination and facialexpression changes. The comparison of these methods applied to different variations in face images offers a comprehensive view of all the subspace methods currently used in face recognition. (ICASSP 03)

Publications:

J. Li, and R. Chellappa, "Structure from planar motion", accepted to IEEE transactions on Image Processing (TIP), Feb., 06

J. Li, S. Zhou and R. Chellappa, "Appearance modeling using a geometric transform", submitted to IEEE transactions on Image Processing (TIP), 2006

J. Li, S. Zhou and R. Chellappa, "Appearance modeling under geometric context", IEEE, International Conference on Computer Vision (ICCV), Oct. 2005

J. Li, and R. Chellappa, "A factorization approach for structure from planar motion", IEEE Workshop on Motion and Video Computing (MOTION), January 5-6, 2005 (oral presentation)

J. Li, R. Chellappa, "Appearance modeling under geometric context for object tracking and recognition", Doctoral Dissertation Proposal

J. Li, S. Zhou, "Probabilistic face recognition with compressed imagery," IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2004

J. Li, S. Zhou and C. Shekhar, "A comparison of subspace analysis for face recognition," ICASSP, 2003 (oral)

The copyright of most papers belongs to IEEE. The copyright of the content of this page belongs to Jian LI.