Face Orientation Prediction Based Facial Landmarks Detection

VWith the development of society and technology,face applications such as face verifi-cation,face recognition and so on increase rapidly.While among those applications,facial landmark detection is one of the most fundamental tasks.Facial landmarks detection(face key points detection,or face alignment)is the task that given a face image,we have to predict the positions of some key points such as eyes,nose,mouth corners etc.Before Deep Learning,cascade regression boosting framework dominates this field.In these days,researchers have use Deep Learning to directly predict the landmarks' position.In this paper,we propose a novel framework combined with convolution neural networks and cascade regression boosting algorithms.First we train a convoltion neural network(CNN)model to classify face orientations(front,left,right),then we train specific LBF feature based cascade regressors for different face oritentation.In addition,we also try to use dataset with 68 landmarks ground truth to predict 5 landmarks.Experiments on several datasets demonstrate the great performance of the proposed framework.Especailly in the task of predicting 5 landmarks,we set a new state-of-the-art mean error record on the dataset of AFW.