Recognition of Image-Orientation-Based Iris Spoofing

Detecting image-orientation iris spoofing and eye laterality using anatomical ocular features and pupil displacement (IEEE TIFS).

Overview

Iris recognition systems rely on fine textural patterns within the iris that are unique to each eye and sensitive to spatial orientation. If an iris image is flipped horizontally or inverted upside-down ($180^\circ$), automated matching algorithms fail to align the patterns, causing false non-matches or allowing orientation-based presentation attack spoofing. In addition, large-scale biometric databases frequently contain metadata errors where eye laterality (left versus right) is mislabeled.

Developed during an International Undergraduate Research Experience Program (REU) at the University of Notre Dame and supported by the WVU FBI Biometric Center of Excellence, this project designed automated methods to classify eye orientation and detect orientation-based iris spoofing before feature extraction.

Technical contributions and methodology

1. Orientation and laterality classification

We designed algorithms to classify the complete orientation state of an iris capture:

  • Distinguishing left from right eye (laterality).
  • Detecting normal right-side-up versus inverted upside-down orientations.

2. Biological ocular feature extraction

We evaluated classical computer vision methods based on anatomical asymmetries of the human eye:

  • Pupillary-to-iris center displacement: The pupil center naturally sits slightly nasal (inward toward the nose) relative to the outer iris boundary. Measuring the horizontal vector between the segmented pupil center and iris center provides a consistent geometric cue for left versus right eye determination.
  • Eyelash distribution: Analyzed directional curvature, thickness, and density of upper and lower eyelashes to resolve vertical orientation (right-side-up versus inverted).

3. Spoof detection and database validation

  • Evaluated on benchmark iris datasets to verify that orientation correction prevents false rejections in verification pipelines.
  • Applied the classification pipeline to detect flipped or inverted images, helping automate quality control and clean metadata errors in biometric databases.

Project details

  • Role: Undergraduate Research Fellow (REU) and Co-Author
  • Affiliation: Computer Vision Research Laboratory (CVRL), University of Notre Dame
  • Collaborators and Co-Authors: Dr. Adam Czajka, Prof. Kevin W. Bowyer, Michael Krumdick
  • Sponsor: West Virginia University (WVU) and FBI Biometric Center of Excellence
  • Timeline: 2015 – 2017
  • Technologies and methods: Classical Computer Vision, Morphological Operations, Pupil and Iris Segmentation, Iris Biometrics, Feature Extraction

Publications

IEEE TIFS 2017

  • Title: Recognition of Image-Orientation Based Iris Spoofing
  • Authors: A. Czajka, K. W. Bowyer, M. Krumdick, and R. G. VidalMata
  • Venue: IEEE Transactions on Information Forensics and Security, Vol. 12, No. 9, pp. 2184–2196, Sept. 2017
  • Links: IEEE Xplore (doi:10.1109/TIFS.2017.2701332)