Department Home

Homepage of Andreas Uhl

Research and projects

Students page of Andreas Uhl

Lab (PS) Topic Description

• A biometric presentation attack detection (PAD) system is typically a binary classifier, discriminating a sample image taken by the sensor from a human body part (real sample, bona fide sample, genuine sample) from a sample image taken by the sensor from a presentation attack instrument (PAI), an artefact used to deceive the system by pretending to be a real / genuine human body part (fake sample, spoofed sample). Thus we have real samples, and fake samples which are used to train the binary PAD classifier.
• Fake samples are difficult to produce, as for each specimen a PAI needs to be created and scanned by the sensor. This is why people have come up with synthetic PAI (fake) samples which are much easier to create. We will work with different types of such synthetic PAI samples and will investigate if these can be used instead of / in addition to fake samples. To summarise, we have three different types of data: (1) Real samples (2) Fake samples and (3) Synthetic samples (trying to mimic fake samples), and the aim was to investigate if type (3) samples can replace type (2) samples as these are much easier to create.
• The major task in this lab is to find potential solutions to shown problems in the following two papers;
  • Andreas Vorderleitner, Andreas Uhl. Finger Vein Spoof GANs: Issues in Presentation Attack Detector Training. In The 40th ACM/SIGAPP Symposium on Applied Computing (SAC ’25), pp. 751-758, Catania, Italy, April 2025, ACM, New York, NY, USA, 2025: paper
  • Moritz Langer, Michael Hafner, Stefan Findenig, Alexandar Radovic, Andreas Vorderleitner, Andreas Uhl. Difficulties in Using Synthetic Data for Presentation Attack Detection in Finger Vein Recognition: The Role of Model Fingerprints. In 2024 IEEE/IAPR International Joint Conference on Biometrics (IJCB), pp. 1-10, 2024.
  In particular, the aim is to experiment with different techniques to remove model fingerprints while maintaining high image quality. In particular, the poor results as exhibited in the above papers should be improved.

First Task

• First we establish the baseline results: the PAD system is trained with type (1) and (2) samples, and also evaluated with those (in a five-fold cross validation protocol). I.e. we start by establishing ground truth for PLUS, IDIAP and SCUT, i.e. the PAD system is trained with type (1) and (2) samples, and also evaluated with those (in a five-fold cross validation protocol). We only use type (1) samples, for which type (2) samples do exist (balanced sets). The folds are constructed by partitioning the users into folds (approximately). We use 4/5 of the data as training data for the k-NN to check the classification accuracy of the remaining 1/5 of the data. This is repeated 5 times (5 folds) to obtain classification rates for the entire datasets.
• Subsequently, we train the PAD sytem with type (1) vs. (3) samples, and evaluate it with type (1) vs. (2) samples (we want to see if training with synthetic data can lead to reliable results to discriminate real from fake samples).
• We use the k-nearest neighbour classifier as PAD binary classifier, different groups use different feature extractions schemes as input to the classifier.
• Perform the evaluation according to the metrics defined in the above papers, ACER, APCER as well as BPCER. Another example paper shows results and also explains data partitioning.

Second Task

• Conduct literature and software research on how to remove model fingerprints (the more techniques you find, the better): can also be disguised as transforming deep fakes to "natural data".
• Conduct literature and software research for deep fake detectors.
• Form groups to implement identified techniques of interest, or create own techniques.
• The data (type (1) and type (2) data) is available at https://www.cosy.sbg.ac.at/~uhl/Data_FV_Spoofing_WS2025_26.zip, also synthetic data created by four distinct GANs.
• Two types of diffusion-based synthetic data for this task: (Residual) Auto Encoder and Variational Auto Encoder

Third Task

• Use deep fake detectors if sample data is (still) recognised as deep fake.
• Check the presence / modification of the model fingerprints in the Fourier / autocorrelation domain.
• Compute image quality metrics, i.e. PSNR and SSIM, to investigate the quality loss.
• Apply methodology of the first task if results are improved.

Dataset description

Each folder contains subfolders for the genuine examples (real samples) and the spoofed ones (fake samples):
genuine
spoofed

Sample naming conventions

The filename are encoded using the following structure: [scanner name]_[DORSAL/PALMAR]_[session ID]_[user ID]_[finger ID]_[image ID].png
an example filename is: PLUS-FV3-Laser_PALMAR_001_01_02_01.png

- scanner name: the name of the scanner used to capture the image, either PLUS-FV3-Laser or PLUS-FV3-LED
- DORSAL/PALMAR: denotes if the image is captured from the palmar or the dorsal side
- session ID: the session ID, two digits, where 01 denotes the first session
- user ID: the user ID, three digits, where 001 denotes the first user and 060 denotes the 60th user
- finger ID: the finger ID, two digits, starting from the left thumb (01) till the right pinky finger (10):
02: left index finger
03: left middle finger
04: left ring finger
07: right index finger
08: right middle finger
09: right ring finger
- image ID: the image ID, two digits, starting from 01.

IDIAP

Samples are stored as follow with the following filename convention: full/bf/004-F/004_L_2. The fields can be interpreted as //-/__. The represents one of two options full or cropped. The images in the full directory contain the full image produced by the sensor. The images in the cropped directory represent pre-cropped region-of-interests (RoI) which can be directly used for feature extraction without region-of-interest detection. We provide both verification and presentation-attack detection protocols for full or cropped versions of the images.

The field may one of bf (bona fide) or pa (presentation attack) and represent the genuiness of the image. Naturally, biometric recognition uses only images of the bf folder for all protocols as indicated below. The is a 3 digits number that stands for the subject's unique identifier. The value can be either M (male) or F (female). The corresponds to the index finger side and can be set to either "R" or "L" ("Right" or "Left"). The corresponds to either the first (1) or the second (2) time the subject interacted with the device.

SCUT

Images are labeled as follow: ID_finger_session_shot_light.bmp, where “ID” stands for client's ID, “finger” ranges from 1 to 6 standing for the index, middle and ring finger of right and left hand respectively, “session” stands for session number which can be "0" or "1" “shot” stands for the considered shot number ranging from 0 to 5, and “light” stands for the level of light intensity which can be an integer between 1 and 6.

Images from the same client are regrouped into a single folder labeled.

NOTE: When processing these images, you only need to consider the first two label, i.e. “ID” and “finger”.