Face recognition

1
Face recognition

• Homero Rios
• Laboratorio Nacional de Informática Avanzada,
  A.C.

2
Contents

• Applications
• Psychophysics Neurophysiological aspects of
  face recognition
• Machine recognition
• Stages of machine recognition preprocessing,
  segmentation, feature extraction, recognition
  classification
• Work at LANIA
• Conclusion

3
Applications

• Credit card, drivers licence, passport
  personal identificaction.
• Mug shot matching
• Bank/store surveillance
• Crowd surveillance
• Expert identification
• Electronic mug shots book
• Electronic lineup
• Reconstruction of face from remains
• Computerized Aging

4
Psychophysics Neurophysiological aspects
offace recognition

• Is face recognition a dedicated process?
• Is face perception the result of holistic or
  feature analysis?
• Ranking of significance of facial features
• Caricatures
• Distinctiveness
• Spatial frequency analysis
• The right hemisphere of the brain
• facial expression
• role of race/gender
• image quality

5
Machine recognition

• Three types of problems
• - recognition of static images ( profiles)
• - recognition from range images
• - recognition from video sequences

6
Machine recognition

• Stages for recognition
• - preprocessing
• - segmentation
• - feature extraction
• - recognition classification

7
Preprocessing

• Enhancements (filtering)
• Normalization (position, rotation, scale, other
  geo. Trans., illumination, etc.)
• Edge detection

8
Segmentation

• Use of controlled background color
• Use of motion
• Contour following and edge grouping
• Active contours
• Top-down search
• Deformable models

9
Feature extraction

• Corner high curvature points
• Template matching
• Deformable models

10
Recognition Classification

• Structural methods (graph matching)
• Statistical methods
• - Eigenfaces, active appearance models
• - Neural nets

11
Assessment of face recognition systems

• Accuracy
• Speed
• Scalability

12
Work at LANIA
13
Work at LANIA
14
Work at LANIA
15
Work at LANIA
16
Conclusion

• There is a tendency to create trainable
  deformable models from examples
• Some models (as active appearance models) combine
  analysis and synthesis of faces
• Only a small amount of knowledge from the
  physchophysics neurophysiological literature
  has been used by automated face recognition
  systems