| With the continuous technological improvement of our country,face recognition technology is developing rapidly.Various face recognition applications have appeared in our daily lives,such as face recognition unlocking on mobile phones,face recognition registration in hotels,facial recognition of train station ticket checking,etc.However,certain security risks have also been exposed during the real world application.Since there are some face recognition systems which not equipped with face anti-spoofing algorithm,criminals can use fake faces of registered users to deceive the face recognition system,such as photos or videos.Therefore,although live face detection has become an important topic in face recognition field and has made certain progress,there are still a number of security issues in practical application scenarios caused by the lack of live face detection function.This paper proposed a face anti-spoofing algorithm based on shallow convolutional neural network.The main advantages of this algorithm are low computing resource consumption and fast computing speed.This paper developed a live face detection platform for access control in laboratory scenarios based on this algorithm.The main functions of the platform are: live face detection,face recognition,access control management,sign in recording,data statistics and visualization and many other functions.In addition,this platform can be connected to a variety of external hardware devices and used in more scenarios,such as pickup in express cabinets by face,etc.,which has practical application value.In order to accomplish the above tasks,this article mainly focuses on the following points:(1)Research and compare the hardware devices used in the current mainstream face recognition field.After a series of performance tests and comparisons,we decided to use a small development board called "Raspberry Pi" as the core of the face recognition hardware device.(2)We proposed a live face detection algorithm based on shallow neural network.This network consists of only two convolutional layers,two pooling layers,and two fully connected layers.The algorithm is tested on datasets such as OULU-NPU.The experimental results show that this algorithm reduces the requirements of computing performance significantly while ensuring that it is available in real world applications,so that live face detection function can be also deployed in low configuration devices.(3)A new live face detection dataset was proposed,which collected real faces,printed face images,fake face videos replayed on 4K ultra-high-definition monitors,and high-definition mobile phone screens under different lighting environments.Compared with the existing data sets,the biggest difference between our method and the existing datasets is that we use an ultrahigh-definition display device as our fake face display device.(4)Taking the face recognition access control in laboratory scenarios as an example,a live face detection platform based on the Raspberry Pi was designed and developed.The main function of the platform is face recognition with live face detection,and it can be linked to a variety of external devices to achieve functions such as access control and sign in recording. |
