Research On Facial Landmark Localization Method Based On Face Detection

Facial landmark localization is a top priority in face recognition and analysis,and it is a prerequisite and a breakthrough for other face-related problems such as automatic face recognition,facial analysis,3D face reconstruction and 3D animation.In recent years,the deep learning method has been successfully applied to many fields such as image recognition and analysis,speech recognition and natural language processing because of its automatic learning and continuous learning ability,and in these areas has brought significant improvement.Therefore,this paper uses the depth learning method to study the location of facial feature points.In the traditional research process,people generally regard face detection and facial landmark localization as two separate issues.But in fact,face detection and facial landmark localization work are closely linked and affect each other.Among them,face detection is the prerequisite and basis of facial landmark localization.In the target face image accurately detect the face area can guarantee the accuracy of the facial landmark localization,but also can narrow the search range,thereby improving the positioning efficiency.Therefore,this paper considers the combination of face detection to improve the accuracy of facial landmark localization.The first combination is the convolution neural network used for face detection cascade with the facial landmark localization convolution neural network before and after,in order to give full play to the role of face detection on the positioning of facial features;The second combination is multitasking.The face detection and facial landmark localization are regarded as two kinds of interrelated tasks.By using the idea of multitasking,a deep convolution neural network model is used to study both tasks so that the two tasks can be mutually promote and reprove each other.Based on this,the main research content and innovation of this paper are:(1)Designed a face detector based on selective search strategy and improvedAlexnet convolution neural network.In this paper,we use the deep learning method and the convolution neural network model for face detection.This is because,the face rectangular box cut out from the input image through the traditional face detection algorithm(such as the most classic Adaboost algorithm),often doping a lot of redundant background,and the face sometimes can not be accurate in the center of the rectangle.The use of convolution neural network can significantly improve these problems,which can be better for facial landmark localization.(2)A facial landmark localization method based on cascaded convolution neural network is proposed.The first level is the face detector mentioned in(1)and the second and third levels are the two-layer convolution neural networks for the coarse location and fine positioning of the facial landmarks.In this paper,the network structure and parameter setting of the two-layer convolution neural network are discussed in detail.(3)A facial landmark localization algorithm based on multi-task deep learning is proposed.In this paper,two multi-task deep learning models are designed: the deep learning model of fusing face detection and facial landmark localization in the middle layer of convolution neural network,and the deep learning model of fusing them in the full connection layer of the convolution neural network.In this paper,we set up multiple sets of control experiments,which based on the Caffe deep learning framework and Python programming.Experimental results show that the above two methods can achieve a higher accuracy compared with the other current good method.