Facial Features Detection: A Comparative Study

Facial feature detection is an essential step in many human face-related applications including facial expression analysis, 3D face reconstruction, pose normalization, vision-based biometrics, and face recognition/verification. Though facial features detection or fiducial facial landmark points may seem an easy process for humans, it faces some challenges as the selection accuracy varies depending on some factors, such as pose, expression, illumination, and occlusion. Several facial landmark detection methods have been introduced to automatically detect those key points over the last ten years. In this paper, a comparative study is conducted to evaluate the efficiency of the state-of-the-art methods used for facial features detection. Seven fiducial points are considered in this study, namely left eye left corner, left eye right corner, right eye left corner, right eye right corner, left mouth corner, right mouth corner, and the tip of the nose. The experiments are conducted on three widely-adopted challenging datasets: BioID, AFLW, and 300W. The experimental results obtained on these three benchmark datasets show that the robustness can be further improved for detection algorithms and reveal future avenues for further research on the topic.