Deformable Iris Matching

A Bayesian approach to deformed pattern matching of iris images

Thornton, Jason, Marios Savvides, and B.V.K. VIjaya Kumar

IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007


title={A Bayesian approach to deformed pattern matching of iris images},

author={Thornton, Jason and Savvides, Marios and Kumar, B.V.K. Vijaya},

journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, volume={29},





We describe a general probabilistic framework for matching patterns that experience in-plane nonlinear deformations, such as iris patterns. Given a pair of images, we derive a maximum a posteriori probability (MAP) estimate of the parameters of the relative deformation between them. Our estimation process accomplishes two things simultaneously: It normalizes for pattern warping and it returns a distortion-tolerant similarity metric which can be used for matching two nonlinearly deformed image patterns. The prior probability of the deformation parameters is specific to the pattern-type and, therefore, should result in more accurate matching than an arbitrary general distribution. We show that the proposed method is very well suited for handling iris biometrics, applying it to two databases of iris images which contain real instances of warped patterns. We demonstrate a significant improvement in matching accuracy using the proposed deformed Bayesian matching methodology. We also show that the additional computation required to estimate the deformation is relatively inexpensive, making it suitable for real-time applications.


An iris pattern exhibiting non-linear deformations as the pupil dilates and constricts. In this work we address the problem of matching such deformed iris patterns.


We model the non-linear iris pattern deformation as local translations of small patches. The basic premise of our model is that, matching iris patterns exhibit coherent translations while non-matching iris patterns exhibit random translations.

Matching Deformed Patches

The local similarities and translations of the iris patches are computed using correlation filters. Correlation filters are a class of classifiers that are designed to estimate the similarity and relative translation between two patterns.

Deformation Prior And Inference

A Gaussian Markov Random Field with hidden states is used to capture a prior distribution over the local translations of the patches along with their occlusion labels.  

Deformed Iris Matching Results

 Here we show an example of iris pattern matches. We show two iris patterns that belong to the same class. The output of our inference process for matching them is shown above. The red regions indicate the occluded regions estimated by our our algorithm. And the green regions are the deformation estimates for the corresponding patches. 

Iris mathcing results on the CMU and CASIA Iris datasets. We compare our algorithm with the standard iriscode matching and optical flow based matching. Our method demonstrates vastly superior performance over the state-of-the-art baselines.

© Carnegie Mellon University Biometrics Center 2015

  • s-twitter
  • s-facebook