Face Detection And Expression Recognition In Natural Environment

Facial expression is the most natural and universal way for human beings to convey their emotional states,and it contains rich emotional information.Automatic facial expression recognition plays a vital role in emotion recognition and has a wide range of application prospects,including human-computer interaction,psychological analysis,etc.Face detection,as the premise and basis of automatic facial expression recognition,has equal importance.With the continuous achievements of artificial intelligence technologies,making use of deep learning to explore effective face detection and expression recognition methods has become a research hotspot in the field of computer vision.In recent years,most face detection and expression recognition methods have used datasets collected in controlled laboratory environment as research objects and achieved outstanding results.However,due to the face images exists complex background,illumination intensity,local occlusion,head deflection,scale variation and other external environmental effects in real natural environment,as well as intra-class differences caused by different individual identities,the existing methods are directly applied to natural environment,the accuracy of face detection and expression recognition is always difficult to obtain the desired results.Aiming at the above problems for face detection and expression recognition in natural environment,the datasets collected in natural environment are used as experimental objects,the corresponding face detection and facial expression recognition methods are researched and constructed respectively.The main work of this paper is as follows:(1)Considering the obvious advantages of YOLOv3 in the field of general object detection,this paper proposes a face detection method based on YOLOv3.In order to adapt it to face detection,softmax is selected as the classification judgment function,the original loss function is appropriately changed,and the dimension of the feature map on the detection layer is reduced.Experimental results on the FDDB dataset show that,compared with Face Boxes and MTCNN,the average accuracy of this method is improved by 1.38% and 2.56% respectively,which provides better face detection performance.(2)Aiming at the problems of facial expression recognition in natural environment,based on generative adversarial networks,this paper proposes a method for facial expression recognition using cross-domain image generation.the Original input image is translated into a hand-drawn image with the same semantics in a cross-domain manner,and then used for facial expression recognition.A corresponding network model,objective function and training algorithm are constructed.Experiments on the RAF-DB dataset show that,compared with AUG-CLOSS and DLP-CNN,the average recognition rate is improved by 1.26% and 2.79% respectively,which proves this method is effective.(3)In order to visually show the actual effect of the proposed face detection and expression recognition method,a corresponding system is designed.The system takes images collected in natural environment as input,and the visual operation of face detection and expression recognition is realized using Electron and Python,which verifies the effectiveness of the proposed methods.It has certain feasibility and applicability when applied to the assessment of concentration for students in online courses.