Xudong Jiang: Research

Research

Ongoing Research Grant

Project Title

Funding Body

Amount of Funding

Grant Period

Role

Biometric and Watermarking Enabled Secure Mobile Media

Thematic Strategic Research Programme Grant No: 062 130 0056

Agency for Science, Technology & Research (A*Star), Singapore

Science and Engineering Research Council

Singapore dollar

854,740.00

2007 – 2010

Principal Investigator

A complete and fully automatic face verification system is built on mobile device - O2 XDA Flame

PhD Students

Dr Manhua Liu (Graduate, 2007)

Dr Bappaditya Mandal (Graduate, 2008)

Dr Wei Yu (Graduate, 2009)

Cong Geng (Waiting for oral defense, 2011)

Chen Tai Pang, Lawrence

Amit Satpathy

Zhenwei Miao

Jianfeng Ren

Jian Lai

Selected Research Projects

Fingerprint Verification, Classification, Retrieval and Identification
Face Detection and Recognition Based on Statistical Estimation
Multimodal Biometric Fusion

The advanced fingerprint recognition includes the robust fingerprint verification and fingerprint retrieval for identification. To apply fingerprint authentication techniques to massive users, we need to increase the applicability of the authentication system. Because a human’s finger is exposed to the outside environment of daily life, a fingerprint sensor may not be able to cope with some extreme skin conditions such as extremely dry and moist skin. Therefore, a fingerprint authentication system will unavoidably encounter a certain amount of low-quality fingerprints. A large number of applications require the system to be robust to low-quality fingerprints. Robust fingerprint recognition improves the system’s reliability in handling low-quality fingerprints, which is crucial for the system’s massive application. Fingerprint retrieval from large database is a crucial part of the automatic fingerprint identification system. Convention exclusive fingerprint classification partitions fingerprints into a few pre-specific human interpretable non-overlapping classes. This limits the efficiency of the fingerprint indexing. The continuous fingerprint classification overcomes the limitation of the number of the classes. However, the exhaustive search of the whole fingerprint database required by this approach could be time consuming. We are exploring method that inherits the merits of both the exclusive and continuous fingerprint classifications and overcomes the limitations and drawbacks of these two conventional approaches. We have established a framework for the fingerprint retrieval from the large database. Our experimental results show that this new approach is promising for fingerprint retrieval from the large database.

Face recognition has attracted many researchers in the area of pattern recognition, machine learning and computer vision because of its immense application potential. Numerous methods have been proposed in the last two decades. However, there are still substantial challenging problems remain to be solved. One of the critical issues is how to extract discriminative and stable features for recognition. Linear subspace analysis has been extensively studied and becomes a popular feature extraction method for face recognition. However, it has still outstanding challenging problems when applied to the face recognition due to the high dimensionality of the face image and the finite number of training samples in practice. We are exploring method of regularizing the statistic learning results to alleviate the problem caused by the finite number of training samples in practice and hence enhance the generalization of the machine learning from face image samples.

Due to vulnerability of the biometric system to environmental noise and variation caused by the user, fusion of several biometric traits is identified as a promising solution. Various fixed rules (min, max, ‘and’, ‘or’, median and mean) and trainable classifiers (linear combination and nonlinear classifiers) are used to combine several biometric traits. We are exploring multimodal biometric fusion by integrating the human expert knowledge into the machine learning process to create more robust fusion with better generalization.