3D face recognition on low-cost depth sensors

This paper deals with the biometric recognition of 3D faces with the emphasis on the low-cost depth sensors; such are Microsoft Kinect and SoftKinetic DS325. The presented approach is based on the score-level fusion of individual recognition units. Each unit processes the input face mesh and produces a curvature, depth, or texture representation. This image representation is further processed by specific Gabor or Gauss-Laguerre complex filter. The absolute response is then projected to lower-dimension representations and the feature vector is thus extracted. Comparison scores of individual recognition units are combined using transformation-based, classifier-based, or density-based score-level fusion. The results suggest that even poor quality low-resolution scans containing holes and noise might be successfully used for recognition in relatively small databases.