  |  |  |  |  |
1. Face Recognition Method
2. Performance Modeling and Prediction of Face Recognition Systems
3. A Near Real-time System for Face, Eye Detection, and Face Recognition
Our face recognition method is along the direction of local matching methods, but has significant differences comparing to the existing algorithms.
We pay close attention to local small patches. The faces are first aligned according to facial feature points and then divided
into a large number of small local patches. A subset of patches along with their sizes and shift window sizes is then selected based on a set of
training samples. The similarities between corresponding patches of different faces are computed, and the classification is the Borda
Count combination of a subset of selected patches.
Publication:
1. Jie Zou, Qiang Ji, and George Nagy, A Comparative Study of Local Matching Approach for Face Recognition, IEEE Transactions on Image Processing,VOL. 16, NO. 10, OCTOBER 2007.
2. Best Paper Award --- Peng Wang, Matthew B. Green, Qiang Ji and James Wayman, "Automatic Eye Detection and Its Validation" , IEEE Workshop on Face Recognition Grand Challenge Experiments (with CVPR), San Diego, CA, June 2005
To predict the success or failure of the face recognition using detected and tracked faces, generic methods are presented to model and predict the face recognition performance.
A performance metric is presented which allows modeling the face recognition performance without using training data. Features are extracted to online predict recognition results
of a set of or individual query data. The presented methods can offline select algorithm parameters to achieve near optimal accuracy under different environments, online predict
recognition performance, and online adjust face alignment for better recognition. The performance of a face recognition system can be greatly improved with the use of the presented methods.
Publications:
1. Peng Wang, Qiang Ji, and James Wayman, Modeling and Predicting Face Recognition System Performance Based on Analysis of Similarity Scores, IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), Vol. 29, No.4, April 2007
2. Peng Wang and Qiang Ji, Performance Modeling and Prediction of Face Recognition Systems, IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2006, New York city.
2. Peng Wang, Lam Cam Tran, and Qiang Ji, Improving Face Recognition by Online Image Alignment, the 18th International Conference on Pattern Recognition (ICPR), Hongkong, August, 2006.
1. Offline system parameters tuning
Using presented performance metric, the system parameters are offline tuned without using any training data. The table below summarizes the recognition accuracy using offline selected parameter and the actual accuracy arrange for different query sets. The accuracy range is obtained from ground truth by exhaustively searching parameters. It shows the the selected parameters without training data can achieve near-optimal accuracy.
2. Improving system accuracy with performance prediction
Using the performance metric, the eye locations are also online updated to further improve recognition accuracy. The eye candidates are searched around the initial eye locations. The eye candidate corresponding to the largest performance metric will be used to align face for recognition. The below table summarizes the recognition accuracy using original and adjusted eyes, based on the initialization of manual and automatic eyes. It shows that the adjusted eyes can provide better accuracy than original eyes.
3.
A Near Real-time System for Face, Eye Detection, and Face
Recognition
We build a near real-time system which
can perform face and eye detection, and online face recognition
simultaneously. The system can run near real-time at a laptop using a simple CCD camera. The
demo video captured from screen is shown as follows. In this demo, the
name of recognized person is shown above his/her face.
|
|
|||||
| Face recognition in
video (for different cameras, lighting and expression) |
||||||