ThirdEye: Triplet Based Iris Recognition without Normalization

Benjamin Fuller; Sohaib Ahmad

arxiv: 1907.06147 · v1 · pith:QJCP67CUnew · submitted 2019-07-13 · 💻 cs.CV

ThirdEye: Triplet Based Iris Recognition without Normalization

Sohaib Ahmad , Benjamin Fuller This is my paper

Pith reviewed 2026-05-24 21:38 UTC · model grok-4.3

classification 💻 cs.CV

keywords iris recognitionnormalizationtriplet lossconvolutional neural networkssegmentationequal error ratebiometrics

0 comments

The pith

Iris recognition achieves competitive accuracy by skipping normalization and applying triplet networks directly to segmented images.

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

The paper tests whether the traditional normalization step after iris segmentation is necessary when using modern deep learning. It presents ThirdEye, a triplet convolutional neural network trained to compare segmented iris images without mapping them to a fixed rectangular region. The system records equal error rates of 1.32 percent, 9.20 percent, and 0.59 percent on the ND-0405, UbirisV2, and IITD datasets. Performance improves over prior work on the most constrained dataset while remaining close on the others. The authors conclude that normalization matters more when imaging conditions are less controlled.

Core claim

ThirdEye directly uses segmented images without normalization and achieves equal error rates of 1.32%, 9.20%, and 0.59% on the ND-0405, UbirisV2, and IITD datasets respectively. For IITD, the most constrained dataset, this improves on the best prior work. However, for ND-0405 and UbirisV2, the equal error rate is slightly worse than prior systems. The concluding hypothesis is that normalization is more important for less constrained environments.

What carries the argument

Triplet convolutional neural network applied directly to segmented iris images to produce embeddings for matching without normalization

If this is right

Normalization can be removed while retaining competitive accuracy on constrained iris datasets.
The importance of normalization increases as imaging conditions become less controlled.
Triplet loss supports direct feature learning from segmented iris regions for biometric comparison.

Where Pith is reading between the lines

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

If segmentation quality is consistently high, other traditional preprocessing steps could be revisited for removal in deep-learning pipelines.
Avoiding normalization would reduce computational overhead in controlled capture settings.
Evaluating the method on datasets with intermediate levels of constraint would map where normalization becomes essential.

Load-bearing premise

The segmentation masks given to ThirdEye have quality comparable to those used in prior normalized systems, so any performance difference can be attributed to the absence of normalization.

What would settle it

Training and testing an otherwise identical normalized pipeline on the exact same segmented images and segmentation masks, then observing whether it matches or exceeds ThirdEye's equal error rate on IITD.

Figures

Figures reproduced from arXiv: 1907.06147 by Benjamin Fuller, Sohaib Ahmad.

**Figure 2.** Figure 2: Segmented iris image. A vanilla triplet network [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Triplet generation. For two arbitrary targets, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Triplet design. The top part of the figure shows training the ResNet with hard triples. The bottom shows feature [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: ROC curve for the ubiris dataset. on the IITD dataset, accuracy is comparable to state of the art without normalization. We next turn to the ND-0405 NIR dataset which is the primary dataset used in this work. The dataset contains some occlusions, motion blur and off-angle images. With the ND-0405 dataset the EER of 1.3% is less than DeepIrisNet2 which again has been trained using normalized images. DeepIr… view at source ↗

**Figure 6.** Figure 6: ROC curve for the ND-0405 and IITD datasets. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

read the original abstract

Most iris recognition pipelines involve three stages: segmenting into iris/non-iris pixels, normalization the iris region to a fixed area, and extracting relevant features for comparison. Given recent advances in deep learning it is prudent to ask which stages are required for accurate iris recognition. Lojez et al. (IWBF 2019) recently concluded that the segmentation stage is still crucial for good accuracy.We ask if normalization is beneficial? Towards answering this question, we develop a new iris recognition system called ThirdEye based on triplet convolutional neural networks (Schroff et al., ICCV 2015). ThirdEye directly uses segmented images without normalization. We observe equal error rates of 1.32%, 9.20%, and 0.59% on the ND-0405, UbirisV2, and IITD datasets respectively. For IITD, the most constrained dataset, this improves on the best prior work. However, for ND-0405 and UbirisV2,our equal error rate is slightly worse than prior systems. Our concluding hypothesis is that normalization is more important for less constrained environments.

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.

Desk Editor's Note private letter to a colleague

ThirdEye gets solid EER numbers on unnormalized segmented irises but the results do not isolate the effect of skipping normalization from possible segmentation differences.

read the letter

The core result is that a triplet-loss CNN reaches EERs of 1.32% on ND-0405, 9.20% on UBIRISv2, and 0.59% on IITD when fed segmented but unnormalized iris images, beating the best prior number on the most constrained set. That is the concrete new data point the paper supplies. The approach is straightforward: take existing segmentation masks, skip the rubber-sheet normalization, and train an embedding network. The authors then observe that the gap versus normalized baselines is smallest on the constrained IITD data and larger on the others, which leads to their hypothesis that normalization matters more when pose and illumination vary. The numbers are reported on standard public datasets and the hypothesis is stated plainly, which is useful for anyone thinking about pipeline simplification. The main weakness is exactly the one flagged in the stress test. The paper gives no quantitative comparison of mask quality (IoU, boundary error, or similar) against the segmentations used in the cited prior work. If the masks supplied to ThirdEye are simply cleaner or more complete, the performance difference could be explained by segmentation rather than the absence of normalization. The abstract and reported experiments do not address this, so the attribution remains open. Training protocol, network details, and any re-implementation of baselines are also absent, which limits how far the numbers can be trusted without further checks. This is the sort of incremental empirical paper that iris biometrics groups would want to see in review so the missing controls can be requested. It is not ready to cite as settled evidence that normalization can be dropped, but the question it asks is worth answering with tighter experiments. I would send it out for refereeing rather than desk-reject.

Referee Report

2 major / 1 minor

Summary. The paper proposes ThirdEye, a triplet CNN iris recognition system that takes segmented iris images as input and skips the standard normalization step. It reports EERs of 1.32% (ND-0405), 9.20% (UBIRISv2), and 0.59% (IITD), with the IITD result claimed to improve on prior work, and concludes that normalization is more important in less constrained environments.

Significance. If the attribution of performance to the lack of normalization can be isolated from segmentation differences, the result would indicate that modern triplet networks can achieve competitive iris recognition without the normalization stage, especially on constrained data. This could simplify pipelines, but the current empirical presentation does not yet establish that isolation.

major comments (2)

[Experimental evaluation / results tables] The central empirical claim (EERs of 1.32/9.20/0.59 % and the hypothesis that normalization matters less on constrained data) requires that the segmentation masks supplied to ThirdEye are at least as accurate as those used by the cited prior normalized systems on the same datasets. No quantitative mask-quality comparison (IoU, boundary error, or equivalent) is provided against the segmentation pipelines of the referenced baselines.
[Methods / experimental setup] The manuscript states concrete EER numbers but supplies no training protocol details, no description of the segmentation method producing the input masks, and no indication that baselines were re-implemented with identical segmentation. Without these, the performance differences cannot be confidently attributed to the removal of normalization.

minor comments (1)

The abstract cites 'Lojez et al. (IWBF 2019)' for the importance of segmentation; the reference list should be verified for completeness and consistency with the full text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the comments identifying gaps in the experimental presentation. We respond point by point below.

read point-by-point responses

Referee: [Experimental evaluation / results tables] The central empirical claim (EERs of 1.32/9.20/0.59 % and the hypothesis that normalization matters less on constrained data) requires that the segmentation masks supplied to ThirdEye are at least as accurate as those used by the cited prior normalized systems on the same datasets. No quantitative mask-quality comparison (IoU, boundary error, or equivalent) is provided against the segmentation pipelines of the referenced baselines.

Authors: We agree that without quantitative mask-quality metrics it is not possible to fully isolate the contribution of omitting normalization from possible segmentation differences. The manuscript contains no such comparison. In revision we will add available details on the segmentation approach used and explicitly discuss this as a limitation of the current evidence for the hypothesis. revision: yes
Referee: [Methods / experimental setup] The manuscript states concrete EER numbers but supplies no training protocol details, no description of the segmentation method producing the input masks, and no indication that baselines were re-implemented with identical segmentation. Without these, the performance differences cannot be confidently attributed to the removal of normalization.

Authors: The manuscript does omit these details. We will revise the methods section to include the training protocol, a description of the segmentation method that produces the input masks, and a clarification that baseline numbers are taken from the published literature rather than re-implementations under identical segmentation. This will make the attribution to normalization removal more transparent. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical measurements on public datasets

full rationale

The paper presents an empirical evaluation of a triplet CNN for iris recognition applied directly to segmented (but unnormalized) images, reporting EER values on ND-0405, UBIRISv2, and IITD. No mathematical derivation, first-principles prediction, or fitted parameter is claimed whose output reduces to its inputs by construction. All load-bearing claims are direct experimental measurements against external public datasets; self-citations are absent from the central argument and the work is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on the domain assumption that triplet loss produces useful iris embeddings from raw segmented pixels and that the supplied segmentations are adequate; no free parameters or invented entities are introduced in the abstract.

axioms (2)

domain assumption Triplet loss produces discriminative embeddings for iris images when trained on segmented data
Invoked by the choice of Schroff et al. architecture and the decision to skip normalization
domain assumption Segmentation quality is comparable across normalized and non-normalized pipelines
Required to attribute performance differences to the removal of normalization

pith-pipeline@v0.9.0 · 5721 in / 1320 out tokens · 51763 ms · 2026-05-24T21:38:48.756967+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages · 5 internal anchors

[1]

Ahmad and B

S. Ahmad and B. Fuller. Unconstrained iris segmentation using convolutional neural networks. In Proceedings of the Asian Conference on Computer Vision. Springer, 2018

work page 2018
[2]

Alonso-Fernandez, P

F. Alonso-Fernandez, P. Tome-Gonzalez, V . Ruiz-Albacete, and J. Ortega-Garcia. Iris recognition based on sift features. In 2009 First IEEE International Conference on Biometrics, Identity and Security (BIdS), pages 1–8. IEEE, 2009

work page 2009
[3]

Arsalan, R

M. Arsalan, R. Naqvi, D. Kim, P. Nguyen, M. Owais, and K. Park. Irisdensenet: Robust iris segmentation using densely connected fully convolutional networks in the images by vis- ible light and near-infrared light camera sensors. Sensors, 18(5):1501, 2018

work page 2018
[4]

K. W. Bowyer and P. J. Flynn. The ND-IRIS-0405 iris image dataset. arXiv preprint arXiv:1606.04853, 2016

work page internal anchor Pith review Pith/arXiv arXiv 2016
[5]

J. Daugman. Iris recognition border-crossing system in the uae. International Airport Review, 8(2), 2004

work page 2004
[6]

J. Daugman. How iris recognition works. In The essential guide to image processing, pages 715–739. Elsevier, 2009

work page 2009
[7]

J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

work page 2009
[8]

Gangwar and A

A. Gangwar and A. Joshi. Deepirisnet: Deep iris represen- tation with applications in iris recognition and cross-sensor iris recognition. In 2016 IEEE International Conference on Image Processing (ICIP), pages 2301–2305. IEEE, 2016

work page 2016
[9]

DeepIrisNet2: Learning Deep-IrisCodes from Scratch for Segmentation-Robust Visible Wavelength and Near Infrared Iris Recognition

A. Gangwar, A. Joshi, P. Joshi, and R. Raghavendra. Deepiris- net2: Learning deep-iriscodes from scratch for segmentation- robust visible wavelength and near infrared iris recognition. arXiv preprint arXiv:1902.05390, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1902
[10]

K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

work page 2016
[11]

In Defense of the Triplet Loss for Person Re-Identification

A. Hermans, L. Beyer, and B. Leibe. In defense of the triplet loss for person re-identiﬁcation. arXiv preprint arXiv:1703.07737, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[12]

Hofbauer, F

H. Hofbauer, F. Alonso-Fernandez, P. Wild, J. Bigun, and A. Uhl. A ground truth for iris segmentation. In 2014 22nd 8 international conference on pattern recognition, pages 527–

work page 2014
[13]

Iris Recognition with Image Segmentation Employing Retrained Off-the-Shelf Deep Neural Networks

D. Kerrigan, M. Trokielewicz, A. Czajka, and K. Bowyer. Iris recognition with image segmentation employing re- trained off-the-shelf deep neural networks. arXiv preprint arXiv:1901.01028, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1901
[14]

Learning-Free Iris Segmentation Revisited: A First Step Toward Fast Volumetric Operation Over Video Samples

J. Kinnison, M. Trokielewicz, C. Carballo, A. Czajka, and W. Scheirer. Learning-free iris segmentation revisited: A ﬁrst step toward fast volumetric operation over video samples. arXiv preprint arXiv:1901.01575, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1901
[15]

Krizhevsky, I

A. Krizhevsky, I. Sutskever, and G. E. Hinton. Imagenet classiﬁcation with deep convolutional neural networks. In Advances in neural information processing systems , pages 1097–1105, 2012

work page 2012
[16]

Kumar and A

A. Kumar and A. Passi. Comparison and combination of iris matchers for reliable personal authentication. Pattern recognition, 43(3):1016–1026, 2010

work page 2010
[17]

LeCun, Y

Y . LeCun, Y . Bengio, and G. Hinton. Deep learning.nature, 521(7553):436, 2015

work page 2015
[18]

Li and H

P. Li and H. Ma. Iris recognition in non-ideal imaging condi- tions. Pattern Recognition Letters, 33(8):1012–1018, 2012

work page 2012
[19]

N. Liu, M. Zhang, H. Li, Z. Sun, and T. Tan. Deepiris: Learn- ing pairwise ﬁlter bank for heterogeneous iris veriﬁcation. Pattern Recognition Letters, 82:154–161, 2016

work page 2016
[20]

Z. Liu, Y . Yin, H. Wang, S. Song, and Q. Li. Finger vein recognition with manifold learning. Journal of Network and Computer Applications, 33(3):275–282, 2010

work page 2010
[21]

Lozej, D

J. Lozej, D. Štepec, V . Štruc, and P. Peer. Inﬂuence of segmen- tation on deep iris recognition performance. In International Workshop on Biometrics and Forensics, 2019

work page 2019
[22]

L. Ma, T. Tan, Y . Wang, and D. Zhang. Efﬁcient iris recogni- tion by characterizing key local variations.IEEE Transactions on image processing, 13(6):739–750, 2004

work page 2004
[23]

Minaee, A

S. Minaee, A. Abdolrashidiy, and Y . Wang. An experimental study of deep convolutional features for iris recognition. In 2016 IEEE signal processing in medicine and biology sympo- sium (SPMB), pages 1–6. IEEE, 2016

work page 2016
[24]

P. R. Nalla and A. Kumar. Toward more accurate iris recog- nition using cross-spectral matching. IEEE transactions on Image processing, 26(1):208–221, 2017

work page 2017
[25]

Nguyen, C

K. Nguyen, C. Fookes, A. Ross, and S. Sridharan. Iris recog- nition with off-the-shelf cnn features: A deep learning per- spective. IEEE Access, 6:18848–18855, 2018

work page 2018
[26]

Othman, B

N. Othman, B. Dorizzi, and S. Garcia-Salicetti. Osiris: An open source iris recognition software. Pattern Recognition Letters, 82:124–131, 2016

work page 2016
[27]

U. Park, A. Ross, and A. K. Jain. Periocular biometrics in the visible spectrum: A feasibility study. In 2009 IEEE 3rd In- ternational Conference on Biometrics: Theory, Applications, and Systems, pages 1–6. IEEE, 2009

work page 2009
[28]

P. J. Phillips, K. W. Bowyer, P. J. Flynn, X. Liu, and W. T. Scruggs. The iris challenge evaluation 2005. In Biometrics: Theory, Applications and Systems, 2008. BTAS 2008. 2nd IEEE International Conference on , pages 1–8. IEEE, 2008

work page 2005
[29]

H. Proenca. Iris recognition: On the segmentation of degraded images acquired in the visible wavelength.IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(8):1502– 1516, 2010

work page 2010
[30]

Proença and L

H. Proença and L. A. Alexandre. Ubiris: A noisy iris image database. In International Conference on Image Analysis and Processing, pages 970–977. Springer, 2005

work page 2005
[31]

Proenca, S

H. Proenca, S. Filipe, R. Santos, J. Oliveira, and L. A. Alexandre. The ubiris. v2: A database of visible wavelength iris images captured on-the-move and at-a-distance. IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(8):1529–1535, 2010

work page 2010
[32]

Proença and J

H. Proença and J. C. Neves. Deep-prwis: Periocular recog- nition without the iris and sclera using deep learning frame- works. IEEE Transactions on Information Forensics and Security, 13(4):888–896, 2018

work page 2018
[33]

K. B. Raja, R. Raghavendra, and C. Busch. Smartphone based robust iris recognition in visible spectrum using clus- tered k-means features. In 2014 IEEE Workshop on Biometric Measurements and Systems for Security and Medical Appli- cations (BIOMS) Proceedings, pages 15–21. IEEE, 2014

work page 2014
[34]

K. B. Raja, R. Raghavendra, V . K. Vemuri, and C. Busch. Smartphone based visible iris recognition using deep sparse ﬁltering. Pattern Recognition Letters, 57:33–42, 2015

work page 2015
[35]

Schroff, D

F. Schroff, D. Kalenichenko, and J. Philbin. Facenet: A uniﬁed embedding for face recognition and clustering. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 815–823, 2015

work page 2015
[36]

Tan and A

C.-W. Tan and A. Kumar. Accurate iris recognition at a distance using stabilized iris encoding and zernike moments phase features. IEEE Transactions on Image Processing , 23(9):3962–3974, 2014

work page 2014
[37]

K. Q. Weinberger and L. K. Saul. Distance metric learning for large margin nearest neighbor classiﬁcation. Journal of Machine Learning Research, 10(Feb):207–244, 2009

work page 2009
[38]

Y . Wen, K. Zhang, Z. Li, and Y . Qiao. A discriminative feature learning approach for deep face recognition. In European conference on computer vision , pages 499–515. Springer, 2016

work page 2016
[39]

R. P. Wildes. Iris recognition: an emerging biometric technol- ogy. Proceedings of the IEEE, 85(9):1348–1363, 1997

work page 1997
[40]

Zhao and A

Z. Zhao and A. Kumar. Towards more accurate iris recogni- tion using deeply learned spatially corresponding features. In Proceedings of the IEEE International Conference on Com- puter Vision, pages 3809–3818, 2017. 9

work page 2017

[1] [1]

Ahmad and B

S. Ahmad and B. Fuller. Unconstrained iris segmentation using convolutional neural networks. In Proceedings of the Asian Conference on Computer Vision. Springer, 2018

work page 2018

[2] [2]

Alonso-Fernandez, P

F. Alonso-Fernandez, P. Tome-Gonzalez, V . Ruiz-Albacete, and J. Ortega-Garcia. Iris recognition based on sift features. In 2009 First IEEE International Conference on Biometrics, Identity and Security (BIdS), pages 1–8. IEEE, 2009

work page 2009

[3] [3]

Arsalan, R

M. Arsalan, R. Naqvi, D. Kim, P. Nguyen, M. Owais, and K. Park. Irisdensenet: Robust iris segmentation using densely connected fully convolutional networks in the images by vis- ible light and near-infrared light camera sensors. Sensors, 18(5):1501, 2018

work page 2018

[4] [4]

K. W. Bowyer and P. J. Flynn. The ND-IRIS-0405 iris image dataset. arXiv preprint arXiv:1606.04853, 2016

work page internal anchor Pith review Pith/arXiv arXiv 2016

[5] [5]

J. Daugman. Iris recognition border-crossing system in the uae. International Airport Review, 8(2), 2004

work page 2004

[6] [6]

J. Daugman. How iris recognition works. In The essential guide to image processing, pages 715–739. Elsevier, 2009

work page 2009

[7] [7]

J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

work page 2009

[8] [8]

Gangwar and A

A. Gangwar and A. Joshi. Deepirisnet: Deep iris represen- tation with applications in iris recognition and cross-sensor iris recognition. In 2016 IEEE International Conference on Image Processing (ICIP), pages 2301–2305. IEEE, 2016

work page 2016

[9] [9]

DeepIrisNet2: Learning Deep-IrisCodes from Scratch for Segmentation-Robust Visible Wavelength and Near Infrared Iris Recognition

A. Gangwar, A. Joshi, P. Joshi, and R. Raghavendra. Deepiris- net2: Learning deep-iriscodes from scratch for segmentation- robust visible wavelength and near infrared iris recognition. arXiv preprint arXiv:1902.05390, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1902

[10] [10]

K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

work page 2016

[11] [11]

In Defense of the Triplet Loss for Person Re-Identification

A. Hermans, L. Beyer, and B. Leibe. In defense of the triplet loss for person re-identiﬁcation. arXiv preprint arXiv:1703.07737, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017

[12] [12]

Hofbauer, F

H. Hofbauer, F. Alonso-Fernandez, P. Wild, J. Bigun, and A. Uhl. A ground truth for iris segmentation. In 2014 22nd 8 international conference on pattern recognition, pages 527–

work page 2014

[13] [13]

Iris Recognition with Image Segmentation Employing Retrained Off-the-Shelf Deep Neural Networks

D. Kerrigan, M. Trokielewicz, A. Czajka, and K. Bowyer. Iris recognition with image segmentation employing re- trained off-the-shelf deep neural networks. arXiv preprint arXiv:1901.01028, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1901

[14] [14]

Learning-Free Iris Segmentation Revisited: A First Step Toward Fast Volumetric Operation Over Video Samples

J. Kinnison, M. Trokielewicz, C. Carballo, A. Czajka, and W. Scheirer. Learning-free iris segmentation revisited: A ﬁrst step toward fast volumetric operation over video samples. arXiv preprint arXiv:1901.01575, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1901

[15] [15]

Krizhevsky, I

A. Krizhevsky, I. Sutskever, and G. E. Hinton. Imagenet classiﬁcation with deep convolutional neural networks. In Advances in neural information processing systems , pages 1097–1105, 2012

work page 2012

[16] [16]

Kumar and A

A. Kumar and A. Passi. Comparison and combination of iris matchers for reliable personal authentication. Pattern recognition, 43(3):1016–1026, 2010

work page 2010

[17] [17]

LeCun, Y

Y . LeCun, Y . Bengio, and G. Hinton. Deep learning.nature, 521(7553):436, 2015

work page 2015

[18] [18]

Li and H

P. Li and H. Ma. Iris recognition in non-ideal imaging condi- tions. Pattern Recognition Letters, 33(8):1012–1018, 2012

work page 2012

[19] [19]

N. Liu, M. Zhang, H. Li, Z. Sun, and T. Tan. Deepiris: Learn- ing pairwise ﬁlter bank for heterogeneous iris veriﬁcation. Pattern Recognition Letters, 82:154–161, 2016

work page 2016

[20] [20]

Z. Liu, Y . Yin, H. Wang, S. Song, and Q. Li. Finger vein recognition with manifold learning. Journal of Network and Computer Applications, 33(3):275–282, 2010

work page 2010

[21] [21]

Lozej, D

J. Lozej, D. Štepec, V . Štruc, and P. Peer. Inﬂuence of segmen- tation on deep iris recognition performance. In International Workshop on Biometrics and Forensics, 2019

work page 2019

[22] [22]

L. Ma, T. Tan, Y . Wang, and D. Zhang. Efﬁcient iris recogni- tion by characterizing key local variations.IEEE Transactions on image processing, 13(6):739–750, 2004

work page 2004

[23] [23]

Minaee, A

S. Minaee, A. Abdolrashidiy, and Y . Wang. An experimental study of deep convolutional features for iris recognition. In 2016 IEEE signal processing in medicine and biology sympo- sium (SPMB), pages 1–6. IEEE, 2016

work page 2016

[24] [24]

P. R. Nalla and A. Kumar. Toward more accurate iris recog- nition using cross-spectral matching. IEEE transactions on Image processing, 26(1):208–221, 2017

work page 2017

[25] [25]

Nguyen, C

K. Nguyen, C. Fookes, A. Ross, and S. Sridharan. Iris recog- nition with off-the-shelf cnn features: A deep learning per- spective. IEEE Access, 6:18848–18855, 2018

work page 2018

[26] [26]

Othman, B

N. Othman, B. Dorizzi, and S. Garcia-Salicetti. Osiris: An open source iris recognition software. Pattern Recognition Letters, 82:124–131, 2016

work page 2016

[27] [27]

U. Park, A. Ross, and A. K. Jain. Periocular biometrics in the visible spectrum: A feasibility study. In 2009 IEEE 3rd In- ternational Conference on Biometrics: Theory, Applications, and Systems, pages 1–6. IEEE, 2009

work page 2009

[28] [28]

P. J. Phillips, K. W. Bowyer, P. J. Flynn, X. Liu, and W. T. Scruggs. The iris challenge evaluation 2005. In Biometrics: Theory, Applications and Systems, 2008. BTAS 2008. 2nd IEEE International Conference on , pages 1–8. IEEE, 2008

work page 2005

[29] [29]

H. Proenca. Iris recognition: On the segmentation of degraded images acquired in the visible wavelength.IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(8):1502– 1516, 2010

work page 2010

[30] [30]

Proença and L

H. Proença and L. A. Alexandre. Ubiris: A noisy iris image database. In International Conference on Image Analysis and Processing, pages 970–977. Springer, 2005

work page 2005

[31] [31]

Proenca, S

H. Proenca, S. Filipe, R. Santos, J. Oliveira, and L. A. Alexandre. The ubiris. v2: A database of visible wavelength iris images captured on-the-move and at-a-distance. IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(8):1529–1535, 2010

work page 2010

[32] [32]

Proença and J

H. Proença and J. C. Neves. Deep-prwis: Periocular recog- nition without the iris and sclera using deep learning frame- works. IEEE Transactions on Information Forensics and Security, 13(4):888–896, 2018

work page 2018

[33] [33]

K. B. Raja, R. Raghavendra, and C. Busch. Smartphone based robust iris recognition in visible spectrum using clus- tered k-means features. In 2014 IEEE Workshop on Biometric Measurements and Systems for Security and Medical Appli- cations (BIOMS) Proceedings, pages 15–21. IEEE, 2014

work page 2014

[34] [34]

K. B. Raja, R. Raghavendra, V . K. Vemuri, and C. Busch. Smartphone based visible iris recognition using deep sparse ﬁltering. Pattern Recognition Letters, 57:33–42, 2015

work page 2015

[35] [35]

Schroff, D

F. Schroff, D. Kalenichenko, and J. Philbin. Facenet: A uniﬁed embedding for face recognition and clustering. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 815–823, 2015

work page 2015

[36] [36]

Tan and A

C.-W. Tan and A. Kumar. Accurate iris recognition at a distance using stabilized iris encoding and zernike moments phase features. IEEE Transactions on Image Processing , 23(9):3962–3974, 2014

work page 2014

[37] [37]

K. Q. Weinberger and L. K. Saul. Distance metric learning for large margin nearest neighbor classiﬁcation. Journal of Machine Learning Research, 10(Feb):207–244, 2009

work page 2009

[38] [38]

Y . Wen, K. Zhang, Z. Li, and Y . Qiao. A discriminative feature learning approach for deep face recognition. In European conference on computer vision , pages 499–515. Springer, 2016

work page 2016

[39] [39]

R. P. Wildes. Iris recognition: an emerging biometric technol- ogy. Proceedings of the IEEE, 85(9):1348–1363, 1997

work page 1997

[40] [40]

Zhao and A

Z. Zhao and A. Kumar. Towards more accurate iris recogni- tion using deeply learned spatially corresponding features. In Proceedings of the IEEE International Conference on Com- puter Vision, pages 3809–3818, 2017. 9

work page 2017