Domain-Specific Human-Inspired Binarized Statistical Image Features for Iris Recognition

Abstract

Binarized statistical image features (BSIF) have been successfully used for texture analysis in many computer vision tasks, including iris recognition and biometric presentation attack detection. One important point is that all applications of BSIF in iris recognition have used the original BSIF filters, which were trained on image patches extracted from natural images. This paper tests the question of whether domain-specific BSIF can give better performance than the default BSIF. The second important point is in the selection of image patches to use in training for BSIF. Can image patches derived from eye-tracking experiments, in which humans perform an iris recognition task, give better performance than random patches? Our results say that (1) domain-specific BSIF features can outperform the default BSIF features, and (2) selecting image patches in a task-specific manner guided by human performance can out-perform selecting random patches. These results are important because BSIF is often regarded as a generic texture tool that does not need any domain adaptation, and human-task-guided selection of patches for training has never (to our knowledge) been done. This paper follows the reproducible research requirements, and the new iris-domain-specific BSIF filters, the patches used in filter training, the database used in testing and the source codes of the designed iris recognition method are made available along with this paper to facilitate applications of this concept. arXiv:1807.05248v2 [cs.CV]

Post-Mortem Iris Recognition—A Survey and Assessment of the State of the Art

“…Until 2016, there were no publicly available databases comprising iris images collected from deceased subjects. Making iris biometrics more robust against post-mortem changes only started in 2018 with proposals for convolutional neural network-based image segmentation [41], [44] and later iris-specific feature representation [45], [49].…”

“…3) Feature Encoding: A variety of feature extraction routines exist for capturing useful textural information from an unwrapped iris. The most common techniques use various image filtering kernels, such as Gabor filters [47], 'sticks' operators [52], or Binarized Statistical Image Features (BSIF) [49], [53], and the filtering result is then binarized to produce an "iris code." 4) Matching: In Daugman's original pipeline for irisbased recognition systems, the matching relies on the "failure of a test of statistical independence."…”

“…• Classical: Iris recognition algorithms in this category are mostly based on Daugman's original pipeline and some improvements that has been suggested over the years, usually proposing other than Gabor filtering kernels.. For instance, in [49] domain-specific (iris-driven) BSIF features were used instead of generic Gabor wavelets to extract textural information from iris images.…”

“…As seen in Figure 5, the best-performing matcher is the combination of a deep learning-based segmentation [41], feature extraction using a traditional hand-crafted method [49] and the matching using the fractional Hamming distance as in conventional iris recognition systems. The second bestperforming matcher employs the same segmentation technique, and employs a mixture of data-driven and Gabor kernels to generate an iris code.…”

“…As above, except for using standard Binary Statistical Image Features (BSIF) as proposed by Kannala and Rathu [66] to extract features. 13) SegNet-based segmentation + human-driven BSIF [41], [49]. Name on plots: SegNet-HumanDrivenBSIF-HD.…”