Protecting privacy while providing online proctoring during an assessment is paramount. With AI face detection, it is essential to understand that YuJa Verity does not recognize faces but only detects whether a face is present.
What is AI Face Detection
AI face detection is the process of using artificial intelligence (AI) algorithms to detect human eyes, eyebrows, nose, and mouth. The AI algorithms are trained on a dataset to help accurately detect faces and reduce false positives.
Understanding Verity's AI Face Detection
Verity's AI face detection is trained on WIDER FACE, a large dataset that provides annotated images and faces. Below are the key highlights of the dataset.
- Diverse Facial Detection
Verity's AI face-detection system has been trained on over 32,000 publicly available images and analyzed more than 393,000 faces, allowing it to detect faces across a wide range of poses, angles, and lighting conditions. - Human-Verified Detection
Before training the AI model, each face was manually analyzed and labeled with a "bounding box," containing a person's forehead, chin, and cheek. - Comprehensive Set of Annotations
The AI is trained in four comprehensive categories:- Scale: The size of the face, measuring from 11 to 300 pixels.
- Occlusion: Determines how obstructed a face is in an image.
- Pose: Recognizes faces that are rotated or turned to the side.
- Events: Trained on over 60 real-world scenarios, from indoor photos to
outdoor parades.
- High Degree of Precision
Post-processing techniques enhance the results by consolidating overlapping detections and eliminating false positives. As a result, the face-detection system achieves a high average precision rate, surpassing 97% accuracy.