Multi-resolution-type AAM for low-resolution images

The purpose of this paper is to extract small faces from an image. From the point of view of integrating the face tracking into a mobile robot, small face tracking is difficult task. When a mobile robot tries to find a person from a far distance, faces on the image is small and its resolution will be degraded largely, which makes face tracking difficult. Therefore, we present the multi-resolution-type active appearance model (AAM) robust to degraded facial images. In the proposed method, the resolution of AAM is reduced adaptively to be as same level as degraded facial images. The accuracy of proposed method is evaluated in terms of shape error between estimated shape by AAM and hand-labeled shape. The simulation results show that multi-resolution-type AAM has the possibility for coping with small faces on the image.