Research On Face Detection Matching And Recognition Algorithm

Face recognition as one of the important technologies of biometric identification technology has become more mature in recent years.With the realization of deep learning,face recognition technology has gradually shifted to the application direction,and has achieved remarkable results in many areas such as security systems,public security systems,and payment systems.Taking into account the complexity of the scene,real-time and accuracy requirements,and the difficulty of effective sample collection,the face recognition system still has room for improvement.To solve face detection and recognition problems,this paper designs a complete face recognition system.The traditional face recognition system based on the combination of traditional feature extraction algorithm and support vector machine and face recognition system based on deep neural network were studied respectively.The robustness of face detection was improved by using multi-task cascaded convolutional neural networks.Face matching recognition is implemented based on a triple-improved deep convolutional network to improve the speed and accuracy of face detection and recognition.Face recognition system includes the research and design of the four major modules,which are image preprocessing,face feature extraction algorithm,face detection algorithm and face recognition algorithm.Image preprocessing and feature extraction are face detection algorithms.,The basic and necessary conditions for face recognition algorithms.Therefore,based on these four modules,the following studies were conducted.First of all,since the color information may be disordered in the actual sample collection process,the image size is not the same,affect the follow-up experiments,so this paper first studies the image preprocessing.Secondly,in the face detection module,this paper studies the SIFT algorithm and LBP algorithm in the traditional method,and makes some improvements to LBP,combines it with SVM,and designs a face detection module based on the traditional algorithm.For the deep learning method,the face detection method based on the improved cascaded convolutional neural network is used to achieve facedetection through multi-task learning three-layer cascaded CNN.Finally,this paper designs face recognition based on the triple-depth algorithm of deep convolutional network optimization,mainly by directly learning the mapping of images to points in the European space,and based on the ternary loss algorithm to determine the characteristics of the two images in the European space.The distance between the points on the two points directly corresponds to whether the two images are similar,so that the face matching recognition is completed.In the experimental verification,the Tensorflow environment under Anaconda3 is built under the Windows 10 system,and the face detection and face recognition based on the traditional algorithm and deep learning algorithm are respectively tested,and based on the LFW face data set and self-made human face.The data set was tested to verify the recognition accuracy of the face recognition system.