Face Detection And Recognition Based On S-CNN

In recent years,with the rapid development of artificial intelligence and deep neural networks,the technology of identity recognition is hotly developed.The main method of identity recognition is to obtain the multimedia information of the identified person by means of cameras,sensors,etc.,and obtain the true identity information of the recognized person through analysis and comparison of the multimedia information.Common identification methods include:face recognition,fingerprint recognition,voice recognition,iris recognition,handwriting recognition,and so on.Among them,the image recognition and processing of the human face has increasingly attracted the attention of researchers.Face detection,facial expression recognition,face recognition,and key point positioning of faces all contribute to rapid progress in face-related tasks.In face detection,the main methods include face detection based on features,face detection based on template matching,face detection based on statistics,the statistical method based on probability,the statistical method based on the support vector machine,and so on.In face recognition,the main methods include a geometrybased method,a subspace-based method,a local feature-based method,and a deep learning-based method.The face detection and face recognition methods used in this article are all based on neural network methods.In the implementation of face detection,this paper uses a cascaded detection network,which is divided into three phases.The network in each phase is the optimization of the network results in the previous phase.The results of its face detection achieved over 95%accuracy on the FDDB dataset.In the realization of face recognition,this paper added a new angle Softmax loss function during the training process,which is different from the ordinary Softmax loss function and can make different facial features in the feature space more obvious,so that the different face identities can be distinguished by the cosine distance.The face recognition network trained by this method obtains an accuracy of 99.42%on the LFW dataset.Finally,this paper applies the network of face detection and face recognition to the face recognition door control system.This paper captures the picture with faces through the camera,transfer the picture to the background server.The server does face detection,face correction and face recognition to the picture.Then it confirms the identity of the person in the picture.And through the data package to inform the microcontroller control relays to switch the door to switch operation.Near the glass door is also placed an ipad to display the identity of the entrants while using voice broadcast welcome.