Face Morphing Attack Detection Based On Multi-frequency Domain Color Information

With the rapid development of today’s network and face recognition technology,the dissemination of digital images is inevitably involved in network life.The characteristics of easy storage,easy dissemination,and easy modification of digital images make them be an important part of multimedia information.As one of the most important biometric information,the face has played a great role in the identification system.Face recognition technology has been widely used in many occasions.From the access control system of entry and exit airports,stations,docks,etc.,to the unlocking software of digital devices such as computers and mobile phones,and even the purchase system of unmanned vending machines,we can see traces of face recognition technology,which gives us daily life brings great convenience.However,these widely used digital images can be easily tampered and forged.On the one hand,the human face is one of the most influential biological features,both from an economic and social perspective;on the other hand,due to its low cost and low-tech features,compared with other biological forms,the spoofing attacks are easier to implement.Recently,more and more investigations and studies have shown that face recognition systems used in practical scenarios are extremely vulnerable to malicious attacks,and attackers can easily deceive existing state-of-the-art commercial face recognition systems by forging faces.Therefore,for the facial image received by the face recognition system,the face forgery attack forensics technology came into being.In digital image forensics technology,face forgery attack detection technology has increasingly become a key part of biometric identification security.Usually,face spoofing attack is a technical means to attack the face recognition system through forged face images,videos,etc.It can tamper the authenticity of the original digital image,forge will the identity of legitimate users,and greatly damage and threaten the security and stability of the face recognition system and even the public order.Therefore,in order to enhance the detection of face spoofing attacks to facial recognition systems,this paper conducts forensics research on the latest face morphing spoofing attacks.Based on the digital image forensics theory,face morphing attack detection technology are studied,and specific main work is summarized as follows:First,a face morphing spoofing attack forensics algorithm based on the difference between high-frequency and low-frequency chromaticity information is proposed.Different from the existing methods of forensics based solely on image texture,this paper takes the difference of high and low frequency chromaticity information between real face and fake(morphed)face images as the starting point.The detection of face fusion attack is achieved by combining the difference of the high-frequency distribution and low-frequency non-uniformity in image chroma information.This paper firstly proposes to convert facial images from RGB color space to HSV color space with richer chroma information and perform chroma enhancement.Secondly,combined with the spatial and frequency domains of the image,Gaussian filter and wavelet transform are used to effectively separate the high-frequency information and lowfrequency information of the image.Then,LBP and color histogram are used to extract the texture features and color features of the image respectively.After feature splicing,the support vector machine is used for classification,which effectively realizes the identification of real faces and fake(morphed)faces.The experimental results show that the performance of the proposed face morphing spoofing attack forensics algorithm is better than that the existing detection methods.Second,a face morphing attack detection algorithm based on noise reduction neural network and mixed-domain image enhancement is proposed.From the perspective of expanding the difference between real and fake images,this paper analyzes the difference between real face and fused face in facial pixels,and fully considers the information in different color spaces.Firstly,the image is enhanced by using Sobel operator,Gaussian filter,homomorphic filter and wavelet transform algorithm based on spatial domain and frequency domain in different color spaces.The denoising neural network DnCNN is used to enhance the high-frequency noise of grayscale image.Then the local binary mode and gray level cooccurrence matrix are used to extract the texture features of the image,which are fused and sent to support vector machine.The real face image and morphing image are classified by support vector machine.Experiments are carried out on two different datasets,FEI_M and HNU_FM.This algorithm is also compared with many existing morphing attack detection methods.The experimental results show that the method has achieved superior detection performance on the above two datasets.