Scholarlink

With the increasing demand for secure access and identification systems, the deployment of face recognition technology has witnessed significant growth. However, the susceptibility to spoofing attacks presents a formidable challenge. Spoofing occurs when unauthorized individuals attempt to access systems using biometric traits that mimic legitimate users. Despite various spoofing detection techniques, a comprehensive solution remains elusive, particularly due to inadequate handling of variations in lighting conditions and image quality. To address this gap, the proposed methodology introduces a novel application of randomized gamma correction, enhancing the model’s robustness by augmenting the dataset with diverse brightness levels. Experiments conducted using a Convolutional Neural Network on the Replay-Attack benchmark dataset validate the effectiveness of this approach, achieving an impressive accuracy of 99.44% and an Equal Error Rate (EER) of 0.73%. The proposed method has a processing time of approximately 0.2254 s per video and features a smaller model size of 8.18 MB compared to existing mainstream neural networks, demonstrating its efficiency.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Face Anti-Spoofing Approach Using Randomized Gamma Correction Based Data Augmentation Technique

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • S. Karthika,
  • G. Padmavathi,
  • R. Bhuvaneshwari

摘要

With the increasing demand for secure access and identification systems, the deployment of face recognition technology has witnessed significant growth. However, the susceptibility to spoofing attacks presents a formidable challenge. Spoofing occurs when unauthorized individuals attempt to access systems using biometric traits that mimic legitimate users. Despite various spoofing detection techniques, a comprehensive solution remains elusive, particularly due to inadequate handling of variations in lighting conditions and image quality. To address this gap, the proposed methodology introduces a novel application of randomized gamma correction, enhancing the model’s robustness by augmenting the dataset with diverse brightness levels. Experiments conducted using a Convolutional Neural Network on the Replay-Attack benchmark dataset validate the effectiveness of this approach, achieving an impressive accuracy of 99.44% and an Equal Error Rate (EER) of 0.73%. The proposed method has a processing time of approximately 0.2254 s per video and features a smaller model size of 8.18 MB compared to existing mainstream neural networks, demonstrating its efficiency.