Geometrical 2D Face Rotation by Using Gabor-tensor-based Active Appearance Model

Abstract: In this paper, we present a novel face rotation methodology, which generates multi-pose faces from one single frontal face image. The proposed method has two major parts: the first part automatically detects twelve initial landmark points, on a given face image for the initialization of the facemask, and the second part applies Gabor-tensors to represent the texture and shape modelling. The proposed method uses both texture and shape information. Texture information is obtained by using spatial frequency in different sizes and orientations. Shape information is obtained by applying the Gaussian model. Experimental results on publicly available face databases show that the proposed approach gives high recognition rates on non-frontal images and it impressively outperforms the performance of other state-of-the-art methods such as active appearance model (AAM), 3D morphable models and their expanded methodologies.

Keywords: Feature, Gabor-tensor, Gaussian analysis, morphological techniques, morphable models, spatial-frequency, shape, texture