The University of Kentucky Center for Visualization & Virtual Environments

Video Rate 3D Data Acquisition

Video Rate 3D Data Acquisition

Investigators:
Laurence Hassebrook


Project Description:
An application of recent interest in image processing is face recognition where research efforts involving 3D scans are becoming increasingly common, but of these efforts that try to make direct comparisons of 3D surfaces and not just compensate for lighting/pose differences in 2D matching algorithms, face/head matching is performed by first mapping the 3D face to a 2D “head shot” depth map that is then processed using some form of Eigenface variant or mean-squared error measure. In many ways, this treatment of face recognition can be thought of as a least squares based approach where faces are compared based on their similarity in only one, the depth, of three directions where two faces are considered perfect matches in the other two, horizontal and vertical, directions. As an alternative to ad hoc surface comparisons, we believe that 3D face recognition can be performed in the Fourier space using the spherical FFT (see Fig. 1 for example basis function). We make this assertion noting that the human head shares many obvious similarities with a sphere, and as such, the (X,Y,Z)-points that make up the data sets of previous research initiatives can be transformed to their spherical coordinates with the origin located at the head’s centroid. See Fig. 2 for example mapping and filtering of head data. Decomposing the head into spherical harmonics instead of some sort of “Eigenheads” has the advantage of being orientation invariant, and therefore, an FFT_{S 2 } face recognition scheme would be immune to the constraints regarding pose.

NOTICE: Some websites to which these materials provide links for the convenience of users are not managed by the University of Kentucky. The University does not review, control, or take responsibility for the contents of those sites. Equal Opportunity/Affirmative Action Employer.