Hybrid close-up model for active face liveness

Face Liveness Detection, also known as Face Anti-Spoofing (FAS), plays a critical role in facial authentication technology by detecting attempts from malicious users to impersonate others or hide their own identity. Within this field, active liveness detection focuses on analyzing both the input signal and user behavior during a designated task to verify the authenticity of the presented face. Although researchers have made substantial progress in FAS, active liveness detection still holds significant potential for improvement. The limited availability of public datasets suitable for active liveness tasks reflects this opportunity, as it often forces researchers to develop and evaluate new solutions using in-house data that cannot be shared due to its sensitive nature, which makes some obtained results irreproducible and significantly limits mutual contribution within the scientific community. In this paper, we introduce a new FAS dataset, UFPR-Close-Up, acquired while users performed close-up movements during active interactions. Instructions for obtaining it for research purposes only are available at https://web.inf.ufpr.br/vri/databases/ufpr-closeup/. The dataset includes genuine samples from volunteer subjects and spoof samples generated using selected face images from the datasets CelebA and CelebV, displayed through several presentation attack instruments. Additionally, we propose a new active spoof detector that combines distortion features with spatial embeddings, achieving an ACER of 4.33% under the full-data protocol and consistently outperforming all recently proposed active liveness models based on similar interaction across all four evaluation protocols of the dataset.