How to accurately segment and match iris images acquired at-a-distance?
Accurate iris segmentation is the key for accurate iris recognition and challenges
when imaging distance is large. Read more on our framework to accurately segment iris
images for at-a-distance iris recognition.
Can we use very low resolution fingerprint images to establish human identity?
Very low-resolution images of about 50 dpi can also reveal useful information for
biometrics. Read more on such
efforts to identify human from level zero features with experimentation on 156
How can we recover better 3D fingerprints using only 3 images?
The photometric stereo recovery under three images can be modelled to achieve more
accurate reconstruction . Read on such efforts in our upcoming publication.
How can we protect biometrics systems from spoof attacks?
Fingerprint devices have shown to be highly vulnerable from artificial
fingerprints. Read more
on our system that detects fake spoof fingerprints from the oxygenated haemoglobin measurements.
How can we recover and match minutiae features in 3D space?
Contactless fingerprint imaging is hygienic and avoids deformations but full
potential from this biometrics is yet to be realized. Read more
on first such efforts to recover 3D fingerprint templates using minutiae features and match
them in 3D space.
How can we recover deeply learned features for advanced iris recognition?
Deep learning capabilities offer tremendous potential to address limitations of currently available iris recognition technologies.
Read more on such efforts to develop deep learning architecture for the iris recognition.
How to ensure higher security at border crossings?
Emerging challenges from sophisticated face masks and make-up pose serious threat
for border security. Our team developed a prototype for automated watchlist
surveillance to assist for law enforcement personnels at border-crossings and some
more details are here.
How unique and stable are our finger knuckle patterns?
Our study on uniqueness of knuckle patterns from over 500 different subjects and
its stability from 6 year interval data provides more insights. More details from
this study in the paper here.
How can we recover 3D features from a single 2D image?
The ordinal measurements under known light source can reveal 3D profile of unknown
surface. Read on such efforts in our recent paper.
How to more accurately and efficiently match 3D palmprints?
Palmprints have been used to establish identity of humans over several decades.
Read more on our investigation into problems with the
state-of-art and successful efforts to develop more accurate/efficient matching
algorithms for palmprints.
PhD students registered elsewhere can explore/contact for collaboration by short (3-6 months) visits
You may contact me for New Projects (MSc Dissertation, Final Year Project) if you are student in PolyU and interested in my research areas
Chengdong Dong*, Ajay Kumar, Eryun Liu, "Think Twice Before Detecting GAN-generated Fake Images from their Spectral Domain Imprints," IEEE International Conference on Computer Vision and Pattern Recognition, CVPR 2022, , June 2022
Kuo Wang* and Ajay Kumar, "Human Identification in Metaverse Using Egocentric Iris Recognition," TechArXiv, DOI: 10.36227/techrxiv.19750411.v1,, pp 1-13, May 2022
Kevin Ho Man Cheng* and Ajay Kumar, "Deep Feature Collaboration for Challenging 3D Finger Knuckle Identification," IEEE Transactions on Information Forensics and Security, pp. 1158 - 1173, September 2021
Hanzhuo Tan* and Ajay Kumar, "Minutiae Attention Network with Reciprocal Distance Loss for Contactless to Contact-based Fingerprint Identification," IEEE Transactions on Information Forensics and Security, pp. 3299 - 3311, April 2021
Kevin Ho Man Cheng* and Ajay Kumar, "Efficient and Accurate 3D Finger Knuckle Matching using Surface Key Points," IEEE Transactions on Image Processing, pp. 8903 - 8915, September 2020