Enhancing Face Anti-spoofing Systems Through Synthetic Image Generation

This study introduces a strategy for synthetic image generation aimed at enhancing the detection capability of facial authentication systems (FAS). By employing various digital manipulation techniques, new synthetic fake images were generated using existing datasets. Through experiments and result analysis, the impact of using these new fake samples on improving the detection accuracy of FAS systems was evaluated. The findings demonstrated the effectiveness of synthetic image generation in augmenting the diversity and complexity of the training data. Fine-tuning using the enhanced datasets significantly improved the detection accuracy across the evaluated FAS systems. Nonetheless, the degree of improvement varied among systems, indicating varying susceptibility to specific types of attacks.