A Region-based Robust 3D Face Recognition

We present a region-based robust 3D facerecognition approach which is robust to facialexpressions, illumination changes andocclusions. Facial surface is often deformed byexpressions. Generally, the mouth is the mostaffected by expressions, whereas the nose is theleast affected and the most static region. For thisreason, we have concentrated on locating thenose tip and segmenting the nose region. Ourmethod can be grouped into two types: Thesurface-based approach which uses curvatureinformation of the face and the statistical-basedapproach which uses subspace analysis. A newalgorithm based on the combination of these twotypes of approaches is presented in this paper.The algorithm extracts the curvature informationof the nose region from range image first, bydecomposing into maximum and minimumcurvature, and then applies PCA (PrincipalComponent Analysis) to reduce the dimension offeature space, respectively. The two features arefused using the sum rule. Our results show thatthe utilization of the cropped nose regionincreases the recognition accuracy up to 96.1percent, where a subset taken from GavabDBdatabase is used to make evaluations