Design And Implementation Of Anti-spoofing Algorithm Based On Embedded Equipment

Face recognition is a kind of biological individual recognition technology based on human face information.In recent years,due to its rapid development and high technology maturity,face recognition technology has been widely used in personal life,mobile payment,public security and justice and many other aspects of social impact.So a lot of researches are used to improve the efficiency and accuracy of face recognition and have achieved fruitful results.However,wide use of face recognition technology has brought great risks,individual identity pretend has made face recognition field of the current problems to be solved.Mask,photos and videos always deceive the face recognition system which enables facial recognition technology in question.Therefore,as an important guarantee for the safe operation of face recognition system,anti-spoofing technology has great research value and social value.At the same time,the current face recognition technology is more widely used in embedded devices such as cell phones and tablet.The limited computing power of embedded devices makes it difficult for it to be faced with the application scenarios with large computational demands.Therefore,improving the execution efficiency of liveness detection algorithm on embedded devices is also a strong support for further improving the face recognition technology.Though the current mainstream in anti-spoofing algorithm is well performed on the dataset,the performance is still poor when it is used on the ground.Therefore,the new goal is to make a new anti-spoofing dataset and a new anti-spoofing algorithm to ensure the accuracy of the algorithm application,and further research is achieved on model compression to improve the performance of the algorithm model on embedded devices.The specific work of this paper is as follows:1.According to the current mainstream in anti-spoofing dataset to analyze the reasons that affect its generalization performance to design and make new anti-spoofing dataset.The specific improvements include: 1.Focus on expanding infrared face dataset and hd RGB dataset;2.Expand the dataset corresponding to left and right Angle of face,pitch Angle and camera distance.2.Aiming at the face attack based on hd video,hd screen and black and white photos,and an anti-spoofing algorithm based on gray value and texture feature is proposed,which is deployed under an infrared camera3.For the attack from HD RGB photos and based on the commonly used Loss function,an improved Loss function algorithm based on decreasing the interval between classes is designed and implemented,which is deployed under the HD RGB camera.4.Based on the algorithm deployment of embedded devices,the basic principle of the Teneary Network is analyzed and the improvement scheme is proposed to reduce the hardware requirements of the actual application of the algorithm model.