Research On Face Recognition Based On Machine Learning Method

Machine learning specializes in how computers learn like human beings,making a significant contribution to the development of face recognition and other artificial intelligent fields.This article explores how to employ some techniques in machine learning to enable computers to better perform face detection and face verification in the field of face recognition.In the task of face detection,we propose a convolutional neural network using feature fusion to detect faces quickly and accurately.First,we quickly get the histogram of oriented gradient(HOG),and then obtain other features from raw pictures using a convolutional neural network called YOLO,which detects multiple objects accurately and quickly.We combine HOG and features extracted by YOLO to form feature map.During the training process,multi-tasks learning and complex sample processing are introduced so that the neural network can locate and classify the object to improve the training results.Then to further improve the performance,we improve the classifier by replacing the fully connected layer with random forest.Evaluation on the dataset FDDB demonstrates that the detection accuracy and speed are improved a lot with the help of the proposed convolutional neural network.When it comes to face verification,we also use the proposed convolutional neural network and modify the loss function,so that the algorithm can authenticate the face image.According to the idea of the nearest neighbor algorithm,this paper uses the ternary loss function,and the similarity between the face images can be better quantified,thus improving the accuracy of face verification.In order to accelerate the convergence of the algorithm when training the algorithm using the ternary loss function,the composition of the triple is optimized.By experimenting on the general data set LFW,it is proved that the method can improve the accuracy of face verification.