Angular-partitioned spin image descriptor for robust 3D facial landmark detection

Spin images have been widely used for representing local threedimensional (3D) shapes in 3D object recognition and 3D facial landmark detection. An improved spin image descriptor is proposed which enhances shape discrimination performance significantly over the spin image. By generating several sub-spin images for angular-partitioned subspaces, the description of unique angular features within local surfaces can be offered. The experimental results show that the proposed angular-partitioned spin image enhances the localisation accuracy of facial landmarks by 47% and detection reliability by 33% as compared to the spin image.