3-D face recognition

2D face recognition:-

Traditional 2D face recognition systems are based on standard photo or video pictures  these are not ground-based measurements and are highly sensitive to changes in ambient lighting or view angle. In addition, they are sensitive to changes in scale, facial accessories (make-up, glasses, beards) and aging of the user.

Facial model for both 2D & 3D

3D face recognition:-

Facial detection

• 3D face recognition is based on anthropometric data – precise measurements of cranial structure and rigid tissues. 
• The 3D recognition  template  is extracted from extremely precise geometric data (sub-millimeter ) about the cranial curvature is those areas where rigid tissues are most visible (i.e eye sockets,chin zone, the bridge of a nose). These areas are the most unique and are unchanging over time, and therefore robust to aging or weight changing in the subject.
• The semantic analysis of the face permits the use of “smart” algorithms for facial camouflage.
• The algorithms recognize that the user has grown a beard, and automatically shifts the emphasis of recognition (matching) to the alternate reliable areas of the face.
•  The same system is used for eye glasses/spectacles, where the algorithms register a non-concave area around the eyes.

Structured light approach to 3D:-

Different features for each face.

• The structured light approach to 3D face recognition, eliminates concerns about poor lighting conditions, structured light 3D technology uses its own light source.
•  Structured light 3D face-readers shine an invisible near infrared pattern of a grid on a user’s face and then maps the geographic pattern of the face based on the distortions it causes in the grid pattern. 

  Procedure:-

  Image acquisition->Pre-processing->Feature Extraction->Classification

  

  Image acquisition:-

  Capture face images & generate 3D models.

  Pre-Processing:

  Normalize images into the same position.
  

  Feature Extraction:-

  Extract the features from normalized face images.
  

  Classification:-

  Design a classifier, train it with dataset, and test its validity.
  
  

  Types of 3D Data:-

  •  Point‐Cloud Representation
  • Range Image
  •  Surface‐normal based
  • Curvature‐Based Representation & 3‐D Voxel Representation

  Sensors:-

  – Passive stereo
  • two cameras with a known geometric relationship are used
  – Pure structured light
  • uses a camera and a light projector with a known geometric relationship.
  A light pattern is projected into the scene, detected in an image acquired
  by the camera
  – hybrid of passive stereo and structured lighting
  • a pattern is projected onto the scene and then imaged by a stereo
  camera rig
  • Even under ideal illumination, it is common for artifacts to occur in
  face regions such as oily regions that appear specular, the eyes, and
  regions of facial hair such as eyebrows, mustache, or beard.
  • Depth of field for sensing (.3m for stereo , 1m for structured)
  • Image acquisition time
  • A 3D shape is illumination invariant
  • Making the 3D image from 2D sensors is not
   
  

  Advantages of 3D face recognition:- 

  • Robustness
  • High accuracy
  • Thoroughness
  • Anti-spoofing
  • Interpolation can be solved 

    Disadvantages of 3d  face recognition:-

  • oily parts of the face with high reflectance may introduce artifacts under certain lighting on the surface.
  • 3D capturing technology requires cooperation from a subject.