Several face recognition methods based on 3D information have been introduced [11-13]. Cartoux et al. [ ] introduced the first contributions in this field, mainly based on extracting a facial curvature representation from distant images and features were used to match the different faces. Xu et al. [48] ​​proposed automatic 3D face recognition by combining global geometric features with local shape variation and resulted in recognition rates of 96.1% and 72.6% when using a gallery of 30 and 120 topics, respectively. Medioni et al. [30] used a variation of the Iterative Closest Point (ICP) algorithm for 3D face recognition and reported a recognition rate of 98% on a gallery of 100 subjects. Huang, Blanz, and Heisele [16] reported a 3D face recognition method that uses a morphable 3D head model to synthesize training images under various conditions. Zhang and Samaras [18] used the Blanz and Vetter morphable model for recognition. The method is presented to work correctly in case of multiple illuminants. The Mahalanobis distance method is used for classification. Basri and Jacobs [17] proposed a technique in which a set of images of a convex Lambertian object obtained under arbitrary illumination can be accurately approximated by a 3D linear space that can be analytically characterized using harmonics spherical surface. Ansari and Abdel-Mottaleb [39] introduced a method based on stereo images, landmarks around eyes, nose and mouth were extracted from 2D images and converted to 3D landmarks. They used the CANDIDE-3 model [40] for facial recognition. A 3D model was obtained by transforming the generic face of CANDIDE-3 to match the landmarks. The eyes, mouth and nose of ...... middle of paper ......s of coefficients. The results obtained by three SVM classifiers are merged to determine the final classification. The experiments were performed on three well-known databases; these are the Georgia Tech Face Database, AT&T “The Database of Faces” and the Essex Grimace Face Database. The work presented in [4] is an extension of [2]. In [4], we used the bit-quantized images to extract features from the curves at five different resolutions. The 15 sets of approximate coefficients are used to train 15 support vector machines and the results are combined by majority vote. In [5] a facial recognition algorithm is proposed to reduce the computational cost of [4]. To reduce the dimensionality of curve features, dimensionality reduction tools such as PCA and LDA methods were evaluated by conducting different experiments on three different databases: ENT, Essex Grimace and Yale Face Database..