ULSee face verification is based on a deep learning algorithm and uses a discriminant classifier to extract facial features, achieving over 98% accuracy on Labeled Faces in the Wild (LFW).
With video-based verification, multiple confident images are selected in one sequence for recognizing faces instead of just a single image. Therefore even with suboptimal facial positioning, the face validator can still perform exceptionally well.
Live detection prevents defeating the verification system with a photo.
- Face recognition has several important modules; we optimize the performance of each to achieve best final accuracy.
Optimal Face Alignment
- Each face is calibrated using highly accurate feature points and pose information.
Solve Face Occlusion
- Overcome partially occluded cases.
- Recognize faces under a wide variety of lighting conditions.
Recognize Pose Variation
- Capture faces from a variety of poses.