SARSpoof: A South Asian Face Anti-Spoofing Dataset

The SARSpoof dataset is a curated collection of face images and videos from the South Asian region (SAR), developed to improve fairness and robustness in face anti-spoofing (FAS) systems. Existing PAD datasets predominantly represent Western and East Asian facial demographics, resulting in accuracy degradation and demographic bias when applied to South Asian or African populations. SARSpoof addresses this gap by providing real, diverse presentation attack samples and bona fide data collected from 40 South Asian subjects, each contributing five unique photos or videos, resulting in a total of 200 bona fide samples. Participants represent multiple regions of South Asia, providing a rich diversity of facial structures, skin tones, and demographic characteristics. To better reflect real-world conditions, the dataset includes samples captured under different lighting environments, as well as images containing various accessories such as glasses, masks, hats, and other natural occlusions. The spoof portion includes realistic replay attacks and other presentation attack modalities captured under practical, uncontrolled settings. Metadata includes device information, environmental context, and optional demographic notes. SARSpoof supports research in biometric security, fairness in AI, presentation attack detection, and dataset-driven bias mitigation. The dataset was created as part of the undergraduate thesis project “Mitigating Ethnic Bias in Face Anti-Spoofing Systems Using MiniFASNet and the SARSpoof Dataset.” Note: To access the dataset, please email [e-mail]. Thank you.