10 - Ear recognition


The ear has a structure that is not random, but well-defined just like the face:

  • the external border (helix);
  • the protrusion (anti-helix) running inside and parallel to the external border;
  • the lobe;
  • the u-shaped socket known as intertragic notch (tacca intratragica) between the entry of the inner ear (meatus) and the lobe

  • less deatils requiring lower resolution
  • static
  • still relatively limited to forensics

Features extraction

Ear localization:

  • Localization of points of interests
    • possible by Neural Networks (trained)
    • select a set of points based on Iannarelli
  • general object recognition
    • the usual matrixes (training set, positive and negative samples‚Ķ)
  • geometric 3d methods
    • reconstruction of the ear using 3D

Ear recognition

  • Iannarelli
    • localize points using Region of Interests
    • 12 measurements are performed starting from helix
    • accuracy is given by the central point
  • Voronoi
    • partition the plane based on closeness to seed points in a specific subset of the plane
    • very sensitive to pose and illumination
  • Gaussian forces
    • all forces around the ear, and they will stretch the ear forming paths