POSE-AWARE FACE SELECTION FOR IMPROVING FACE IDENTIFICATION RELIABILITY

Аннотация

Face identification systems are widely used in access control, attendance monitoring, surveillance, and human–
computer interaction applications. Although modern FaceID systems based on deep facial embeddings achieve high recognition
accuracy under controlled conditions, their reliability decreases when faces are captured at non-frontal angles.
In real-world environments, camera position and person movement often cause significant yaw and pitch variations,
reducing the visibility of facial landmarks and distorting the extracted facial features. This paper investigates the influence
of face pose angles on FaceID matching performance. The effect of yaw and pitch angles on facial embedding similarity
was experimentally evaluated using cosine distance as the matching metric. A pose-dependent score matrix was constructed
to analyze how the matching score changes under different horizontal and vertical face orientations. Based on
the obtained results, a pose-aware face image selection algorithm was proposed. According to this algorithm, face images
are accepted for identification when the horizontal angle between the camera and the human face is within 0°–57°, and
the vertical angle is within −56.25° to +56.25°. The experimental results showed that excluding highly deviated face
images before identification improves the reliability of the FaceID system. The proposed pose-aware filtering approach
reduced false identification cases by an average of 22%. These results confirm that face pose control is an important
preprocessing stage for improving the practical reliability of FaceID systems in real-world camera-based applications