Recognizing Faces In 3D Images Even In Presence Of Occlusions

We have implemented a full automatic pipeline that is able to automatically detect, normalize and recognize the faces in presence of extraneous objects. The face detector uses curvature analysis, surface registration through ICP and Gappy PCA classification. After detection and normalization, the face images are restored using an occlusion detection algorithm followed by Gappy PCA reconstruction. Feature extraction and matching are performed by Fisherfaces recognition, but any state of the art approach may be applied. The proposed system has been tested using the 951 acquisitions from the UND database processed with an artificial occlusions generator. Realworld objects have been used to hide part of the faces. The results show that the approach can be adopted to improve the robustness of 3D face recognition systems in case of occlusions with an extension up to 30% of the face image.