Finding matching Features on Shapes

• When do two points on different shapes/scans represent the same feature?

• Check if surrounding regions are similar

• We need feature descriptor that summarizes the surrounding region

Classical descriptors: Curvature #

• Differential featrures describe characteristics of surrounding surface
• But: differential features are noisy on meshes and captured data

Classical descriptors: Spin Images #

• Spin a virtual paper around the normals which collects all vertices that pass through it

• The resulting spin images can then be used to compare possible feature positions

• Doesn’t work so well, if feature is at the border of shape, where information is missing, because spin images will look really different

Classical Descriptors: Point Feature Histogram #

• For a point find its neighbors

• Compute histogram from tupes of

  • based on distances,
  • angular distance to normals,
  • optionally curvature etc

• Doesn’t work so well, if feature is at the border of shape, where information is missing, because histograms will look really different