Recognition of 3D Faces with Missing Parts based on Profile Networks

3D face recognition solutions proposed so far compare face scans of a probe subject against archived face scans (gallery), assuming the entire scans are available for both probe and gallery. Though the assumption is reasonable for subjects in the gallery---gallery scans are typically acquired in controlled environments following a collaborative protocol---the same assumption is not general if applied to probes. In fact, when face recognition is performed in un-collaborative contexts, the acquired probes may correspond to just a part of the face or may be altered by occlusions. In this work, we explicitly address these difficulties and propose and experiment an original solution to 3D face recognition that is inherently capable to accomplish the recognition task even in cases where the probe scan represents just a part of the face. In the proposed approach, distinguishing traits of the face are captured by first extracting SIFT keypoints on the face scan and then measuring how the face changes along linear paths, namely profiles, built between pairs of keypoints. The approach is experimented using the Face Recognition Grand Challenge dataset.