Facial Landmark Detection Algorithm in Complex Scenes

Nowadays, Convolutional neural networks (CNNs) have become the benchmark technology for many computer vision applications. Face landmark detection has been an important topic over the last few decades. However, most algorithms are designed for small to medium-sized poses of faces and lack accuracy in complex scenes such as lighting, large poses, and occlusion. There are some challenges: first, the appearance of the face varies more from the frontal view to the profile view. Second, human faces are easily affected by occlusions or illumination, so it is hard to provide the most appropriate location. In this article, we proposed using joint CNNs to improve the accuracy of facial landmark estimation. Extensive experiments conducted on large angles and severe occlusion challenging databases such as Menpo and COFW have also demonstrated the superiority of our proposed method in challenging scenarios.