| With the development of the information age,machine learning has been widely studied.The face detection and recognition technology,based on deep learning algorithm and convolutional neural network(CNN),has been rapidly developed.When extracting facial feature vectors,the traditional face detection and recognition algorithm is vulnerable to the external environment and other variables,which makes it difficult to obtain the face information and complete the face matching and recognition.In the existing face detection and recognition algorithms,there are relatively few studies on the variation factors,such as illumination intensity,posture rotation and partial occlusion in face detection.So when the above-mentioned change factors appear,the accuracy of real-time face detection and recognition will be reduced.Therefore,under the consideration of the changing factors,how to improve the face detection ability and the accuracy of face recognition effectively is a major research topic.Based on the convolutional neural network model,a multi-continuous convolutional neural network model is proposed for face detection firstly,to reduce the influence of changing factors,such as face pose rotation,illumination intensity and partial occlusion.Then the improved convolutional neural network model is designed and trained.Compared with the training effect of Facenet model,the network model with better effect is selected to design the face recognition module.Finally,based on the designed algorithm,the design of student attendance system is completed and tested.The main specific work of this thesis is as follows:1.According to the changing factors in face detection,a multi-level convolutional neural network model is proposed.The model is for detecting the face in the real-time picture,and the face prediction is performed layer by layer through multiple sets of convolutional neural networks.On the one hand,this method can reduce the influence of changing factors,such as illumination intensity and face posture rotation,on the accuracy of face detection.On the other hand,by calculating the average of the face image prediction results by multi-group convolutional neural networks as the final face detection result,the accuracy of face detection is greatly improved.Compared with the detection effect of the Haar-based AdaBoost algorithm,the results show that the multi-convolution network model has better actual detection results.2.Based on the Facenet deep CNN,this thesis uses the Inception-resnet-V1 network structure,selects the Triplet loss ternary target loss function,and combines the CASIA-FaceV5 face dataset to carry out the training experiment of Facenet deep convolutional neural network model.The experiment found that the Facenet deep CNN model has a good training effect on the small sample face database,and can recognize the face information effectively.3.Based on the CNN structure,this thesis designs an improved convolutional neural network model,further increases the number of network layers,adjusts and optimizes the network structure,aiming at effectively extracting facial features and improving the recognition accuracy of face recognition.In order to verify the effectiveness of the algorithm,this paper uses the CASIA-FaceV5 face database to carry out training experiments on the improved CNN model.Compared with the existing algorithms,the experimental results show that the model has a certain improvement in face recognition accuracy.When faced with a face database with a small sample size,it has a good application effect.4.Based on the designed algorithm,combined with the camera and MySQL database,the design and test of the student attendance system was completed.The MySQL database is used to store picture information,including student number,name,gender,major,and facial features.The experimental results show that the student attendance system based on face detection and recognition has good real-time and feasibility. |
