Pose-normalized View Synthesis of a Symmetric Object Using a Single Image

Object recognition under varying pose is a challenging problem, especially when illumination variations are also present. In this paper, we propose a pose-normalized view synthesis method under varying illuminations for symmetric objects. For a given non-frontal view of a symmetric object under non-frontal illumination, the mirror image of the original view is equivalent to the view when the object rotates around the Y -axis by the same angle as the original view but in the opposite direction, and under opposite illumination condition in the X direction. Exploiting the bilateral symmetry of the object, we generate the mirror view of the object under the same illumination condition as the original view on a pixel-by-pixel basis. The frontal view under the same illumination is then easily obtained using view morphing techniques.