Automatic landmarking of 3D images

Finding the intrinsic coordinate system

The first step in our automatic landmarking scheme is to estimate an intrinsic coordinate system of faces scanned in 3D. For this we have developed a method based on the horizontal symmetry. This coordinate system provides an excellent basis for subsequent landmark positioning and model-based refinement such as Active Shape Models, outperforming other landmark localisation methods including the commonly-used ICP+ASM approach. [1]

These four images show the order of events :-

1. Firstly the points of local curvature extrema are located in the shape data:
2. Then the x-axis is located by looking for parallel pairs of points:
3. This identifies the plane of symmetry, in which the y-axis can be defined by the line ab:
4. Finally the z-axis is defined as the cross-product of the other two axes:

Locating the landmarks

Once the coordinate system is established, a reference face, with known landmarks, is used to initialise the landmark positions. Then the method of Ruiz et al. [2,3] is applied to fit the landmarks to the new face.

Comparison of automatic and manual landmarking:

[1] WP Koppen, CH Chan, WJ Christmas and J Kittler An intrinsic coordinate system for 3D face registration In Proc ICPR, 2012.

[2] MC Ruiz and J Illingworth Automatic landmarking of faces in 3D - ALF3D In IEE International Conference on Visual Information Engineering, 2008.

[3] MC Ruiz Automatic face landmarking in 3D PhD thesis, 2011.