Automatic 3D face recovery from a single frame of a RGB-D sensor

Three dimensional (3D) facial analysis is becoming increasingly important for various tasks in computer vision, human-computer interaction and computer graphics. An automatic approach for obtaining a complete and accurate 3D facial model is crucial for those applications such as 3D video conference and personalized 3D facial animations. This paper presents a novel approach to automatically reconstruct a complete 3D facial model. The proposed method only utilizes a single shot of a RGBD camera regardless of noises and data missing from the input, such as holes on the surface due to self-occlusion. The 3D face models can be reconstructed automatically without any manual intervention. The experiments demonstrate the fine-grained results with both online captured RGB-D data and 3D scanned database.