Fingerprint Image Restoration Using the U-Net Deep Learning Model
(2026) In Master's Theses in Mathematical Sciences FMAM05 20252Mathematics (Faculty of Engineering)
- Abstract
- This thesis investigated the use of convolutional neural networks (CNNs) for image restoration of fingerprint images, focusing on the reconstruction from raw sensor images to denoised fingerprint images, suitable for matching. Fingerprint images from sensors often introduce noise and artifacts that inhibit the matching process. The goal was to develop a CNN-based approach capable of removing artifacts such as moiré patters that degrade fingerprint quality, thereby restoring the fingerprint image. A U-Net architecture was used as a baseline model and extended with several architectural modifications such as Convolutional Block Attention Modules (CBAM), gated skip connections and dilated convolutions. In addition, the effects of different... (More)
- This thesis investigated the use of convolutional neural networks (CNNs) for image restoration of fingerprint images, focusing on the reconstruction from raw sensor images to denoised fingerprint images, suitable for matching. Fingerprint images from sensors often introduce noise and artifacts that inhibit the matching process. The goal was to develop a CNN-based approach capable of removing artifacts such as moiré patters that degrade fingerprint quality, thereby restoring the fingerprint image. A U-Net architecture was used as a baseline model and extended with several architectural modifications such as Convolutional Block Attention Modules (CBAM), gated skip connections and dilated convolutions. In addition, the effects of different batch sizes and learning rates for the ADAM optimizer were evaluated. The models were trained with synthetic fingerprint data, including a dataset augmented with image transformations to increase the training dataset size. The results show that CNN-based restoration can surpass traditional ISP pipelines, especially when training and test domains are well aligned. Learning rate selection held significant importance, with 10⁻⁴ consistently yielding the lowest False Reject Rates (FRR). Data augmentation improved robustness and frequently reduced FRR relative to identical models trained on non-augmented data. However, generalization to different datasets remained limited, highlighting the need for a training dataset representative of diverse data. The most promising model architecture for restoration was U-Net, which is the simplest of all the proposed models. Since the training dataset was very limited in size, the risk of overfitting was high, and it seemed that the more complex models tended to overfit relatively quickly. The simple U-Net, in contrast, seemed to generalize the best, and had more consistent performance on the testing data than every other model. Overall, it seems that CNN-based restoration methods offers a promising approach to reconstruct fingerprints from raw sensor images, and that their role and performance could be increased with further development. (Less)
- Popular Abstract (Swedish)
- När biometriska metoder används för autentisering är det viktigt både att processen är korrekt, och går snabbt. Givetvis ska en obehörig användare inte kunna passera autentiseringsprocessen, och givetvis är det fördelaktigt om autentiseringen inte tar lång tid. Vid biometrisk fingeravtrycksigenkänning finns flertalet faktorer som försvårar matchningen. Brist på ljus kan göra fingeravtrycksbilden för mörk, och därmed ha bristande information. När fingret trycks mot skärmen kan artefakter som moirémönster upkomma. Moirémönster är ett slags interferensmönster som bildas av överlappet mellan displayens pixelmönster och sensorns elektrodmönster.
Medan traditionella matchnings-metoder fungerar bra, är de ofta specifika till särskilda typer av... (More) - När biometriska metoder används för autentisering är det viktigt både att processen är korrekt, och går snabbt. Givetvis ska en obehörig användare inte kunna passera autentiseringsprocessen, och givetvis är det fördelaktigt om autentiseringen inte tar lång tid. Vid biometrisk fingeravtrycksigenkänning finns flertalet faktorer som försvårar matchningen. Brist på ljus kan göra fingeravtrycksbilden för mörk, och därmed ha bristande information. När fingret trycks mot skärmen kan artefakter som moirémönster upkomma. Moirémönster är ett slags interferensmönster som bildas av överlappet mellan displayens pixelmönster och sensorns elektrodmönster.
Medan traditionella matchnings-metoder fungerar bra, är de ofta specifika till särskilda typer av sensorer och användarfall. Således är de inte generella nog att fungera bra under nya förhållanden, utan kräver anpassningar beroende på fall. En stor del av dessa anpassningar tar form av förbearbetning av fingeravtrycksbilderna, innan matchningen utförs. Under arbetet har traditionella bearbetningsmetoder jämförts med metoder baserade på neurala nätverk. Som grund för arbetet har en modell kallad U-Net använts, och flera utbyggnader av detta nätverk har utvärderats.
Att träna neurala nätverk att bearbeta och restaurera bilder på fingeravtryck är oerhört utmanande. Det finns en stor variation av artefakter och förhållanden som påverkar bildens kvalitet. Till exempel ger ett "normalt" fingeravtryck en väldigt annorlunda bild än ett fuktigt fingeravtryck. Fokus i det här arbetet har varit på normala fingeravtryck. Även då är det utmanande att utveckla stabila, bra modeller. Variationer som huruvida bilden är ljus eller mörk, lågupplöst eller högupplöst, eller om fingret täcker sensorn eller inte, gör det svårt att träna en alltäckande modell.
Genom att använda den matchningsmetod som finns på Precise Biometrics var det enkelt att jämföra olika modellers kvalitet. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9220891
- author
- Heurlin de Oliveira, Albert LU and Stenbäcken, Elin LU
- supervisor
- organization
- course
- FMAM05 20252
- year
- 2026
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Convolutional Neural Network, Neural Network, Fingerprint Image Restoration, AI, Deep Learning, Image Analysis, U-Net, CNN, Moire Patterns, Biometrics, Biometric Security
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- 2026:E2
- ISSN
- 1404-6342
- other publication id
- LUTFMA-3602-2026
- language
- English
- id
- 9220891
- date added to LUP
- 2026-02-10 10:37:25
- date last changed
- 2026-02-10 10:37:25
@misc{9220891,
abstract = {{This thesis investigated the use of convolutional neural networks (CNNs) for image restoration of fingerprint images, focusing on the reconstruction from raw sensor images to denoised fingerprint images, suitable for matching. Fingerprint images from sensors often introduce noise and artifacts that inhibit the matching process. The goal was to develop a CNN-based approach capable of removing artifacts such as moiré patters that degrade fingerprint quality, thereby restoring the fingerprint image. A U-Net architecture was used as a baseline model and extended with several architectural modifications such as Convolutional Block Attention Modules (CBAM), gated skip connections and dilated convolutions. In addition, the effects of different batch sizes and learning rates for the ADAM optimizer were evaluated. The models were trained with synthetic fingerprint data, including a dataset augmented with image transformations to increase the training dataset size. The results show that CNN-based restoration can surpass traditional ISP pipelines, especially when training and test domains are well aligned. Learning rate selection held significant importance, with 10⁻⁴ consistently yielding the lowest False Reject Rates (FRR). Data augmentation improved robustness and frequently reduced FRR relative to identical models trained on non-augmented data. However, generalization to different datasets remained limited, highlighting the need for a training dataset representative of diverse data. The most promising model architecture for restoration was U-Net, which is the simplest of all the proposed models. Since the training dataset was very limited in size, the risk of overfitting was high, and it seemed that the more complex models tended to overfit relatively quickly. The simple U-Net, in contrast, seemed to generalize the best, and had more consistent performance on the testing data than every other model. Overall, it seems that CNN-based restoration methods offers a promising approach to reconstruct fingerprints from raw sensor images, and that their role and performance could be increased with further development.}},
author = {{Heurlin de Oliveira, Albert and Stenbäcken, Elin}},
issn = {{1404-6342}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master's Theses in Mathematical Sciences}},
title = {{Fingerprint Image Restoration Using the U-Net Deep Learning Model}},
year = {{2026}},
}