Face spoofing detection: theory and practice
summary
Covers practical pipeline: early detectors use LBP+SVM and color histograms or image‑quality metrics to spot print or screen artifacts, while modern systems use transfer learning and CNNs such as ResNet50/DeepPix trained on larger but often imbalanced datasets. Production deployment combines multi‑shot active liveness (blink, mouth, movement prompts) with single‑frame checks, runs in a web app that captures images, and yields reliable authentication in practice but still fails on high‑quality 3D masks and varied cameras/displays.
AI-written summary · about
chapters (17)
- 0:00Introduction and speaker background
- 1:10Problem definition and spoof attack types
- 3:18Common face spoofing attack methods
- 5:49Basic approaches for spoof detection
- 8:19Single-shot versus multi-shot detection methods
- 9:26Performance metrics for evaluation
- 10:51Historical datasets in spoof detection
- 11:58Pre-deep learning methods and techniques
- 13:28Pattern-based detection with LBP
- 14:30Color distribution evaluation methods
- 16:37Image quality metrics for detection
- 17:41Rise of deep learning in spoof detection
- 18:47Early deep learning attempts and breakthroughs
- 20:36Future trends in multi-shot detection
- 22:47Practical challenges and real-world use
- 23:28Q&A: Handling downloaded photos and deepfakes
- 25:59Q&A: Applications in banking and authentication
related talks
08mAdvancements in facial recognition
26mVirtual U: Defeating Face Liveness Detection by Building Virtual Models...
26mNew Face, Who Dis? Recent Adversarial Approaches to Facial Recognition
32mHeart to Heart and Eye to Eye with Deep Fakes
16mThe Threat of Deepfakes to Remote Identity Verification Systems
28mML solutions for the real world problems of identity verification
23mMultimodal Video Face Liveness: A Better Alternative?
58mPushing Deepfakes to the Limit-Fake video calls with AI