The Research On Face Morphing Attacks Detection Based On Feature Cascade

In this era of rapid development of the information security technology,face recognition,as an important biometric technology,has been widely used for banks,hotels,transportation,and other fields for identity verification.Recently,a variety of attacks against face recognition systems were appeared,among which face morphing attack posed a serious threat to the security of the existing face recognition systems.With the deepening of research,various typical face morphing attack detection algorithms emerge as the times require.Among them,deep learningbased methods have become the current research hot spot,but the traditional convolutional neural network often has the problems of a large amount of calculation and many parameters.Algorithms based on lightweight networks can be applied to resource-constrained application scenarios,and the network is also widely used in image classification,face recognition and semantic segmentation.Aiming at the current face morphing attack,this paper studies the structure design and network optimization of lightweight network based on convolutional neural network and face morphing theory.The corresponding optimization methods for face morphing attack detection are as follows:Based on the analysis of patch-level features and edge features of face morphing image,a face morphing attack detection method based on patch-level features and lightweight network is proposed.We design a new lightweight network structure,which can learn the local features of human face,so as to enhance the data set and extract the discrimination information.It utilizes the combination of three "Blocks" structures for learning and adds a more lightweight attention mechanism module of ECA-Net to the "inverted residual" structure,so that the network can maintain the detection accuracy while reducing the parameters.Then "two-level"classification model is used.The first level classification is to output the probability value of local face combined with lightweight network,and the second level classification is to integrate all patch level features of face for recognition.Experimental results and analysis show this method can significantly improve the detection accuracy of face morphing attacks.In order to make full use of the characteristics of convolutional layers and enhance the semantic information of shallow features and the resolution of deep features,this paper proposes a lightweight face morphing attack detection algorithm based on two-channel attention mechanism.It takes the lightweight network structure MobileNetV2 as the benchmark network architecture,and CS-DSCA attention model is designed independently.First,the CA channel attention and spatial attention are cascaded into the input and the CA channel attention unit can fully extract network features,and while Atrous Convolution is used to increase the receptive field.Then the lightweight network is used to discriminate and analyze the face morphing attack.The experimental results and analysis prove that the method proposed in this paper can improve the accuracy of face morphing attack detection to a certain extent,and the proposed network model also has high computational efficiency,which can effectively improve the security of the face recognition system.