An Adaptive Face Model for Real-Time Eyes and Mouth Tracking

We introduce an adaptive face model to be used with extended template matching (ETM) technique [1] to track eyes and mouth in real-time. Although ETM is shown to be robust for these facial components detection, its' performance reduces when the face is not facing forward. This is due to failure of determining correct face center during the tracking, especially from non-frontal pose. Our proposed adaptive face model is meant to resolve this problem. It is simple and reliable to track these facial components within approximately 70 degrees toward left and right. By using face geometrical relationships between both eyes, we create a face model adaptively by dividing the detected face region into right and left regions based on discriminant analysis method. This is done continuously in each frame to create the adaptive face model. It works as the base model during facial components searching using ETM. Results reveal that the proposed method performs better than [1] and tracks at averaged rate of 30 fps.