pith. sign in

arxiv: 1907.09380 · v1 · pith:5GMZ35WSnew · submitted 2019-07-22 · 💻 cs.CV · cs.LG

DeepIris: Iris Recognition Using A Deep Learning Approach

Shervin Minaee , AmirAli Abdolrashidi This is my paper

Pith reviewed 2026-05-24 18:16 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords iris recognitiondeep learningresidual CNNbiometricsfeature learningend-to-end modelvisualization
0
0 comments X

The pith

An end-to-end residual CNN jointly learns iris features from few images per class and performs recognition.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a residual convolutional neural network that trains end-to-end to extract features from iris images and classify them at the same time. It uses a standard iris dataset with only a few examples per identity and reports gains over earlier methods. The authors also describe a visualization step that marks the iris regions most responsible for each decision. If the approach holds, biometric pipelines could become simpler by removing separate feature-engineering stages. The same structure is presented as applicable to other recognition tasks that need scalable training.

Core claim

The central claim is that a residual CNN framework jointly learns feature representation and performs iris recognition when trained on few images per class from a well-known dataset, yielding promising results and improvements over previous approaches, while a visualization technique identifies the iris areas that most influence the recognition outcome.

What carries the argument

residual convolutional neural network that jointly learns feature representation and performs recognition

If this is right

  • The model improves recognition accuracy over prior methods on the tested dataset.
  • The same end-to-end structure applies to other biometrics recognition tasks.
  • The framework supports more scalable and accurate biometric systems by reducing data needs.
  • The visualization step reveals iris regions that drive the classification decisions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • An end-to-end model could remove the need for hand-crafted feature extractors in deployed iris systems.
  • If the visualization consistently points to the same iris zones, those zones could guide future sensor design or image cropping rules.
  • Training with few samples per class may allow faster adaptation when new identities are added to an operational database.

Load-bearing premise

The chosen iris dataset and the limited number of training images per class are representative enough for the network to learn features that generalize beyond the training conditions.

What would settle it

Run the trained model on iris images from a second dataset captured under different lighting or sensor conditions and measure whether accuracy falls below the reported gains.

Figures

Figures reproduced from arXiv: 1907.09380 by AmirAli Abdolrashidi, Shervin Minaee.

Figure 1
Figure 1. Figure 1: The images from the first (on top) and second layers of scattering [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The residual block used in ResNet Model To perform recognition on our iris dataset, we fine-tuned a ResNet model with 50 layers on the augmented training set. The overall block-diagram of the ResNet50 model, and how it is used for iris recognition is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The architecture of ResNet50 neural network [13], and how it is transferred for iris recognition. The last layer is changed to match the number of [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Six sample iris images from IIT Delhi dataset [30]. [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The saliency map of important regions for Iris recognition. [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
read the original abstract

Iris recognition has been an active research area during last few decades, because of its wide applications in security, from airports to homeland security border control. Different features and algorithms have been proposed for iris recognition in the past. In this paper, we propose an end-to-end deep learning framework for iris recognition based on residual convolutional neural network (CNN), which can jointly learn the feature representation and perform recognition. We train our model on a well-known iris recognition dataset using only a few training images from each class, and show promising results and improvements over previous approaches. We also present a visualization technique which is able to detect the important areas in iris images which can mostly impact the recognition results. We believe this framework can be widely used for other biometrics recognition tasks, helping to have a more scalable and accurate systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes DeepIris, an end-to-end residual CNN framework for iris recognition that jointly learns feature representations and performs classification. It is trained on a well-known iris dataset using only a few images per class, claims promising results with improvements over prior approaches, and introduces a visualization technique to highlight salient iris regions. The work suggests the framework could extend to other biometric tasks for more scalable systems.

Significance. If the empirical claims hold under proper evaluation, the approach could demonstrate the viability of residual CNNs for iris recognition with limited per-class samples and add interpretability via visualization. This would be relevant to biometrics applications, but the absence of any reported metrics, baselines, or robustness tests in the manuscript description prevents assessment of whether the results actually advance the state of the art or generalize beyond the training distribution.

major comments (2)
  1. [Abstract] Abstract: The central claim that the model 'show[s] promising results and improvements over previous approaches' is unsupported because the manuscript supplies no quantitative metrics (e.g., recognition accuracy, EER, or ROC curves), no description of baselines, no error bars, and no evaluation protocol. This absence is load-bearing for the empirical contribution.
  2. [Abstract] Abstract: Training is restricted to 'only a few training images from each class' on a single well-known dataset, yet no cross-dataset, cross-sensor, or robustness experiments (e.g., against occlusion, off-angle views, or illumination changes) are described. Without these, the assertion of a 'more scalable' framework cannot be evaluated.
minor comments (1)
  1. [Abstract] The visualization technique is mentioned but not described in sufficient detail (e.g., which layer activations are used or how saliency maps are generated), limiting reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the major comments point by point below. We agree that the empirical claims in the abstract require quantitative support and additional experiments to be properly evaluated, and we will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the model 'show[s] promising results and improvements over previous approaches' is unsupported because the manuscript supplies no quantitative metrics (e.g., recognition accuracy, EER, or ROC curves), no description of baselines, no error bars, and no evaluation protocol. This absence is load-bearing for the empirical contribution.

    Authors: We agree that the current manuscript does not supply the quantitative metrics, baselines, error bars, or evaluation protocol needed to support the abstract's claims. We will revise the manuscript to add a full experimental section reporting recognition accuracy, EER, ROC curves, baseline comparisons, error bars from repeated runs, and a clear evaluation protocol on the standard dataset. The abstract will be updated to include specific results rather than unsupported claims. revision: yes

  2. Referee: [Abstract] Abstract: Training is restricted to 'only a few training images from each class' on a single well-known dataset, yet no cross-dataset, cross-sensor, or robustness experiments (e.g., against occlusion, off-angle views, or illumination changes) are described. Without these, the assertion of a 'more scalable' framework cannot be evaluated.

    Authors: We agree that the absence of cross-dataset, cross-sensor, and robustness experiments limits evaluation of the scalability claim. The present work demonstrates feasibility on one standard dataset with limited samples per class. In revision we will add robustness tests (e.g., occlusion and illumination variations) using available data where feasible, discuss limitations on generalization, and moderate the abstract and conclusion claims about scalability. revision: partial

Circularity Check

0 steps flagged

No circularity: purely empirical DL training with no derivations or self-referential predictions

full rationale

The paper describes training a residual CNN end-to-end on a standard iris dataset and reporting recognition accuracy; it contains no equations, no first-principles derivations, no fitted parameters renamed as predictions, and no load-bearing self-citations that reduce the central claim to its own inputs. All reported results follow directly from the training procedure on the chosen data split, which is externally verifiable and does not rely on any internal redefinition or uniqueness theorem imported from the authors' prior work. This is the normal, non-circular case for an applied computer-vision paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the empirical performance of a standard residual CNN trained with limited per-class data; no new mathematical axioms or invented physical entities are introduced.

pith-pipeline@v0.9.0 · 5663 in / 1039 out tokens · 18237 ms · 2026-05-24T18:16:00.112487+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · 2 internal anchors

  1. [1]

    ”A survey on antispoofing schemes for fingerprint recognition systems.” ACM Computing Surveys (CSUR) 47.2 (2015): 28

    Marasco, Emanuela, and Arun Ross. ”A survey on antispoofing schemes for fingerprint recognition systems.” ACM Computing Surveys (CSUR) 47.2 (2015): 28

    work page 2015

  2. [2]

    ”Highly accurate palmprint recognition using statistical and wavelet features.” Signal Processing and Signal Processing Education Workshop (SP/SPE), IEEE, 2015

    Minaee, Shervin, and AmirAli Abdolrashidi. ”Highly accurate palmprint recognition using statistical and wavelet features.” Signal Processing and Signal Processing Education Workshop (SP/SPE), IEEE, 2015

    work page 2015

  3. [3]

    Burge, eds

    Bowyer, Kevin W., and Mark J. Burge, eds. ”Handbook of iris recognition”. London, UK: Springer, 2016

    work page 2016

  4. [4]

    ”Robust face recognition via multimodal deep face representation.” IEEE Transactions on Multimedia 17.11 (2015): 2049-2058

    Ding, Changxing, and Dacheng Tao. ”Robust face recognition via multimodal deep face representation.” IEEE Transactions on Multimedia 17.11 (2015): 2049-2058

    work page 2015

  5. [5]

    ”Face recognition using scattering convolutional network.” Signal Processing in Medicine and Biology Symposium (SPMB)

    S Minaee, A Abdolrashidi, and Y Wang. ”Face recognition using scattering convolutional network.” Signal Processing in Medicine and Biology Symposium (SPMB). IEEE, 2017

    work page 2017

  6. [6]

    Kumar and A

    A. Kumar and A. Passi, Comparison and combination of iris matchers for reliable personal authentication, Pattern Recogni- tion, vol. 43, no. 3, pp. 1016-1026, Mar. 2010

    work page 2010

  7. [7]

    Farouk, Iris recognition based on elastic graph matching and Gabor wavelets, Computer Vision and Image Understanding, Elsevier, 115.8: 1239-1244, 2011

    RM. Farouk, Iris recognition based on elastic graph matching and Gabor wavelets, Computer Vision and Image Understanding, Elsevier, 115.8: 1239-1244, 2011

    work page 2011

  8. [8]

    Belcher and Y

    C. Belcher and Y . Du, Region-based SIFT approach to iris recognition, Optics and Lasers in Engineering, Elsevier 47.1: 139-147, 2009

    work page 2009

  9. [9]

    ”Iris recognition us- ing multiscale morphologic features.” Pattern Recognition Letters 65: 67-74, 2015

    S Umer, BC Dhara, and Bhabatosh Chanda. ”Iris recognition us- ing multiscale morphologic features.” Pattern Recognition Letters 65: 67-74, 2015

    work page 2015

  10. [10]

    ”Iris recognition using scattering transform and textural features.” Signal Processing and Signal Processing Education Workshop (SP/SPE), IEEE, 2015

    S Minaee, A Abdolrashidi, and Y Wang. ”Iris recognition using scattering transform and textural features.” Signal Processing and Signal Processing Education Workshop (SP/SPE), IEEE, 2015

    work page 2015

  11. [11]

    ”Gradient-based learning applied to docu- ment recognition.” Proceedings of the IEEE: 2278-2324, 1998

    LeCun, Yann, et al. ”Gradient-based learning applied to docu- ment recognition.” Proceedings of the IEEE: 2278-2324, 1998

    work page 1998

  12. [12]

    A Krizhevsky, I Sutskever, GE Hinton, ”Imagenet classification with deep convolutional neural networks”, Advances in neural information processing systems, 2012

    work page 2012

  13. [13]

    ”Deep residual learning for image recog- nition.” Proceedings of the IEEE conference on computer vision and pattern recognition

    He, Kaiming, et al. ”Deep residual learning for image recog- nition.” Proceedings of the IEEE conference on computer vision and pattern recognition. 2016

    work page 2016

  14. [14]

    ”Segnet: A deep convolutional encoder-decoder architecture for image segmentation.” IEEE transactions on pattern analysis and machine intelligence 39.12: 2481-2495, 2017

    Badrinarayanan, Vijay, Alex Kendall, and Roberto Cipolla. ”Segnet: A deep convolutional encoder-decoder architecture for image segmentation.” IEEE transactions on pattern analysis and machine intelligence 39.12: 2481-2495, 2017

    work page 2017

  15. [15]

    Faster r-cnn: Towards real-time object detection with region proposal networks

    Ren, S., He, K., Girshick, R., Sun, J. “Faster r-cnn: Towards real-time object detection with region proposal networks”, In Advances in neural information processing systems, 2015

    work page 2015

  16. [16]

    ”Learning a deep convolutional network for image super-resolution.” European conference on computer vision

    Dong, Chao, et al. ”Learning a deep convolutional network for image super-resolution.” European conference on computer vision. Springer, Cham, 2014

    work page 2014

  17. [17]

    Deep-Emotion: Facial Expression Recognition Using Attentional Convolutional Network

    Minaee, Shervin, and Amirali Abdolrashidi. ”Deep-Emotion: Facial Expression Recognition Using Attentional Convolutional Network.” arXiv preprint arXiv:1902.01019, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1902

  18. [18]

    ”Deep learning face representation by joint identification-verification.”, NIPS, 2014

    Sun, Yi, et al. ”Deep learning face representation by joint identification-verification.”, NIPS, 2014

    work page 2014

  19. [19]

    ”MTBI Identification From Diffusion MR Images Using Bag of Adversarial Visual Features.” IEEE transactions on medical imaging, 2019

    Minaee, Shervin, et al. ”MTBI Identification From Diffusion MR Images Using Bag of Adversarial Visual Features.” IEEE transactions on medical imaging, 2019

    work page 2019

  20. [20]

    ”A deep unsupervised learning approach toward MTBI identification using diffusion MRI.” Engineering in Medicine and Biology Society (EMBC), IEEE, 2018

    Minaee, Shervin, et al. ”A deep unsupervised learning approach toward MTBI identification using diffusion MRI.” Engineering in Medicine and Biology Society (EMBC), IEEE, 2018

    work page 2018

  21. [21]

    ”Convolutional neural networks for sentence clas- sification.”, Conference on Empirical Methods on Natural Lan- guage Processing, 2014

    Kim, Yoon. ”Convolutional neural networks for sentence clas- sification.”, Conference on Empirical Methods on Natural Lan- guage Processing, 2014

    work page 2014

  22. [22]

    ”Learning to rank short text pairs with convolutional deep neural networks.”, SIGIR conference on research and development in information retrieval, ACM, 2015

    A Severyn, A Moschitti. ”Learning to rank short text pairs with convolutional deep neural networks.”, SIGIR conference on research and development in information retrieval, ACM, 2015

    work page 2015

  23. [23]

    ”Automatic question-answering using a deep similarity neural network.” Global Conference on Signal and Information Processing, IEEE, 2017

    S Minaee, Z Liu. ”Automatic question-answering using a deep similarity neural network.” Global Conference on Signal and Information Processing, IEEE, 2017

    work page 2017

  24. [24]

    Neural Machine Translation by Jointly Learning to Align and Translate

    Bahdanau, Dzmitry, Kyunghyun Cho, and Yoshua Bengio. ”Neural machine translation by jointly learning to align and translate.” arXiv preprint arXiv:1409.0473 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  25. [25]

    ”CNN features off-the-shelf: an astounding baseline for recognition.” IEEE conference on computer vision and pattern recognition workshops, 2014

    AS Razavian, H Azizpour, et al. ”CNN features off-the-shelf: an astounding baseline for recognition.” IEEE conference on computer vision and pattern recognition workshops, 2014

    work page 2014

  26. [26]

    ”An experimental study of deep convolutional features for iris recognition.” signal process- ing in medicine and biology symposium (SPMB), IEEE, 2016

    S Minaee, A Abdolrashidi, Y Wang. ”An experimental study of deep convolutional features for iris recognition.” signal process- ing in medicine and biology symposium (SPMB), IEEE, 2016

    work page 2016

  27. [27]

    ”Imagenet: A large-scale hierarchical image database.” 2009 IEEE conference on computer vision and pattern recognition

    Deng, Jia, et al. ”Imagenet: A large-scale hierarchical image database.” 2009 IEEE conference on computer vision and pattern recognition. IEEE, 2009

    work page 2009

  28. [28]

    https://pytorch.org/

    work page

  29. [29]

    Ajay Kumar and Arun Passi, ”Comparison and combination of iris matchers for reliable personal authentication, Pattern Recognition, vol. 43, no. 3, pp. 1016-1026, Mar. 2010

    work page 2010

  30. [30]

    https://www4.comp.polyu.edu.hk/ csajaykr/IITD/Database- Iris.htm

    work page

  31. [31]

    ”Visualizing and understanding convo- lutional networks.” European conference on computer vision, springer, Cham, 2014

    M Zeiler, R Fergus. ”Visualizing and understanding convo- lutional networks.” European conference on computer vision, springer, Cham, 2014

    work page 2014