nent

Available online xxxx of views available for training the models. The idea is to construct models only from frontal and profile For example, Gong et al. [1] represented faces by either normalized intensities or using the composite face representation scheme based on the Gabor wavelet transform. The authors investigated the possibility of identifying facial pose using the facial manifold. They have shown that images corresponding to pose changes of a classifier instead of SVM and adding one extra class to represent non-faces. Another SVM-based pose estimation strategy can be found in Li et al. [6]. A pose differentiation by k-means clustering was proposed by Lee et al. [7]. Okada et al. [8,9] proposed a model, coined PCMAP. It computes bidirectional mappings between facial images and physical parameters (3D head rotation angles), via parameterized manifold representations of faces in the PCA sub-space. The model is subsequently used for view-independent face recognition. Finally, in a recent publication by Sanderson et al. [10], non-frontal views are artificially synthesized from the fronta