A Strong Start to the Year for the Department of Audiovisual Analysis and Biometric Systems

At WACV, the team will present, among others, studies by Joanna Gajewska, Alicja Martinek and Ewelina Bartuzi‑Trokielewicz, focusing on the effectiveness of synthetic‑speech detectors in real fraud scenarios and the analysis of manipulated political advertisements based on lip‑sync techniques. The researchers also prepared a unique dataset of 224 manipulated videos paired with their original counterparts, enabling an in‑depth examination of how such manipulations affect content semantics and biometric system performance.

Another major accomplishment is the development of two new datasets — “Read and Tell” and “VIBEFACE.” These were created collaboratively by the entire team. The datasets support the evaluation of biometric systems in realistic conditions, taking into account demographic diversity, speaking style, recording quality and operational constraints.

At ACM WWW, Ewelina Bartuzi‑Trokielewicz and Alicja Martinek will present the results of long‑term analyses of fraudulent advertising campaigns powered by generative AI. Their research demonstrates how complex, dynamic and difficult to detect modern advertising fraud ecosystems have become, involving thousands of ads and hundreds of campaigns. The researchers reconstructed the structure of these operations, examining their dynamics, scaling strategies, masking techniques and resource rotation.

The team will also showcase two studies related to the APAKT project. Michał Koźbiał and Mateusz Kowalczyk developed an age‑classification method based on the analysis of sexual dimorphism, body shape and facial features, improving accuracy and robustness against adversarial attacks. Additionally, Michał Koźbiał will present an innovative approach to age estimation based on body silhouettes and inter‑person relationships within an image. Importantly, this method was tested on real Dyżurnet.pl reports and significantly reduces misclassification errors between minors and adults.

This strong start to the year demonstrates that the Department of Audiovisual Analysis and Biometric Systems is developing technologies with real impact on digital security — from combating deepfakes, to advancing biometric methods, to analyzing large‑scale fraud and disinformation campaigns.