Shade Face: Multiple image-based 3D face recognition

Three-dimensional face recognition is illumination invariant, however the acquisition process itself is not. In active 3D recognition, multiple images are captured while the face is actively illuminated with different patterns. We propose a 3D face recognition paradigm that bypasses reconstruction and exploits the plethora of information available in multiple images of a person acquired while varying the illumination. Illumination is varied by scanning a horizontal and then a vertical white stripe on the computer screen in front of the subject. Subtracting ambient light leaves images illuminated by the screen from different angles. The contourlet coefficients of the images are calculated at different scales and orientations and then projected to PCA subspace to remove redundancy. The subspace contourlet coefficients of multiple images are stacked to form a global face representation. Sliding windows are used during matching to remove the disparity between the face locations with respect to the screen. The proposed algorithm was tested under varying ambient conditions and compared to a known 3D face recognition technique. Verification results on data from the same subjects show the strength of our algorithm.