Face Identification
1. Objectives
- Automatic recognition for access control
- Remote control (no access card)
- Reduction of database search
2. Hypotheses
- Good record conditions: light, background, distance
- Cooperation: in front of camera, eyes opened, mouth closed
- Human supervisor for doubtful cases
3. Methods
Global Comparison (Neural Network):
- Too much memory needs and time consumption
- Too large learning time
- Discrimination between local and global information ?
Face Normalization and Comparison:
- Reference points are difficult (visible, resolution)
- Important details are lost in front of large areas
- Incomplete normalization (2D image from 3D scene)
Features:
- Extract normalized features (grey level, distance)
- Secure features difficult to find
- Database is small (a few features by face)
4. Feature extraction
Computation of the edges (large grey transitions)
Extraction of the head and its contour
Vertical and horizontal references based on grey level profile
-> Compute normalized distances between lines
5. Matching
- Compute features of images of database (done once)
- Compute features of new image
- Search best match (quick distance estimation)
6. Problems
- Light conditions have a large influence on detectability
- Allowable rotation is limited (but validity is checked)
- Image resolution is not enough to get details
- Classification is still trivial
7. Other Topics