Hui Song
MS Computer Science

Face Recognition via Image Synthesis

My goal is to automatically recognize a face in a photograph or in a video frame. To do this, I first generate an image (using 3D computer graphics) that matches a photographic image.

I use 3D laser scanner to capture the surface geometry of a face. Then I photograph the face from different directions and under different lighting conditions. I project a white disk onto CSIT's 8-foot by 16-foot Power Wall as the light source, then synchronize the shutter on a camera with the light source via network. The direction of the light and the direction of the view are all that are needed to reconstruct the reflectance function at points on the face.

I cluster the reflectance functions to match just a few basic regions of the face (such as lips, cheeks, forehead). Then I apply these clustered reflectances to the polygonal mesh, apply photon mapping with subsurface scattering, and generate the realistic 3D image. This is compared to the photograph for a possible match.

Courses taken outside of major
STA 5106 Computational Methods in Statistics I
STA 5107 Computational Methods in Statistics II