Facial Landmark Localization With Cacaded Convolutional Neural Network

Facial landmark localization is the key algorithm of face detection and recognition prob-lem. We present a new approach for estimating the positions of facial keypoints with a carefully designed cascade convolution neural network.At each layer in the network, sub-convolution neural networks are trained to localize keypoints area or locating high accuracy keypoints.We estimate two rectangular areas in the image with diffirent sizes. Then one area is trained to localize the positions of eyes, nose, mouth, and the other one is trained to localize the positions of facial contour.Extensive experiments show that our approach works well for occlusions, large pose variations and extreme lightings.