[Truncated abstract] Biometrics has been an active research area due to its enormous potential applications in video surveillance, human-machine interaction and access control systems. Among the biometric traits, the human face is the most publicly accepted biometric because of its non-intrusiveness and easy data acquisition. Most of the work on face recognition has been accomplished using 2D data. 2D face recognition systems are not robust to variations in pose, illumination conditions and facial expressions. With the rapid advancements in the development of data capturing technologies (e.g. Minolta Vivid and Microsoft Kinect), the acquisition of 3D data is becoming a more feasible task. 3D data processing has the potential to overcome the limitations and drawbacks faced by 2D facial data. Most of the existing 3D face recognition systems rely on the surface registration of the gallery and probe faces and/or on complex feature matching techniques. These methods are sensitive to facial expression and computationally expensive and are not suitable for real-world applications. In this thesis, we present novel algorithms based on low-level geometrical signatures which can be extracted at a low computational cost. To address the issue of facial expression variations, we adopt various machine learning techniques. This thesis is organized as a set of papers published in journals or currently under review. Three different local geometric feature based approaches have been proposed and their efficiency has been demonstrated through extensive experimental evaluations on the largest publicly available 3D face datasets. First, a fast and fully automatic approach based on four kinds of low-level geometrical features collected from the semi-rigid facial regions was devised and used to represent 3D faces. As a result, the effects of the deformed facial regions are avoided. The extracted features revealed to be efficient in computation and robust in the presence of facial expressions. A region-based histogram descriptor computed from these features was used as a single feature vector for a 3D face. The resulting feature vectors are independent of the coordinate system and hence can be tolerant to minor pose variations. A Support Vector Machine (SVM) was then trained as a classifier based on the proposed histogram descriptors to recognize any test face. In order to combine the contributions of the two semi-rigid facial regions (eyesforehead and nose), both feature-level and score-level fusion schemes are tested and compared. The experimental results demonstrate that feature-level fusion achieves a higher performance compared to score-level fusion. Second, in order to further increase the computational efficiency and robustness, a computationally efficient 3D face recognition approach is presented based on a novel facial signature called Angular Radial Signature (ARS). This approach extracts a set of ARS features from the semi-rigid regions of a 3D face. It was demonstrated that the extraction of these signatures is highly efficient (low computational cost). The Kernel Principal Component Analysis (KPCA) is subsequently used to extract the mid-level features from the ARSs to achieve a greater discriminative power and to deal with the linearly inseparable classification problem...