Research On Digital Image Forensics Technology Of Spoofing Attack In Face Recognition

With the development of face recognition technology,it has become a trend to use face biometric information for identity authentication.Face recognition has been widely used in a series of scenes closely related to border entrance management,crime monitoring and investigation,and financial security protection,which brings great convenience to our daily life.However,more and more research shows that existing face recognition systems are extremely vulnerable to malicious face spoofing attack,and attackers can easily fool the most advanced commercial face recognition systems by fake faces.Therefore,face spoofing attack forensics,which attempts to verify the authenticity and integrity of face images or videos collected by face recognition systems,emerges at the right moment and has become an important research hotspot in the field of biometric recognition security.Face spoofing attack usually takes place outside the face recognition system.By using forged face images,video and other ways,it can easily succeed in the imitation of legitimate user’s identity.This has greatly damaged the interests of legitimate users and seriously threatened the security of face recognition systems,and even threatened public security and social stability.Therefore,in order to improve the ability of existing face recognition systems against face spoofing attack,in this paper,we research on two typical face spoofing attacks,which are face photo/video attack and face morphing attack,by utilizing traditional digital image forensics theory and focusing on the trace/clue left by different spoofing attacks for its forensics analysis.The main work and contributions of this dissertation are summarized as follows:Firstly,a face spoofing detection scheme based on color texture Markov feature and support vector machine recursive feature elimination is proposed.From the perspective of recaptured image,the adjacent facial pixels discrepancy between the real and fake face is analyzed,and texture information between color channels is fully considered.Firstly,the directional difference filter is used to capture the facial texture difference between the real and the fake face,which can be regarded as low-level features of color texture Markov feature.Then,the facial texture difference is modeled by the Markov process to form a high-level representation of the low-level features.Experiments on four public benchmark databases indicate that the proposed scheme can effectively resist photo/video spoofing attack in face recognition.Secondly,a face morphing spoofing attack detection method based on camera sensor pattern noise is proposed.Different from the previous texture feature based face morphing detection methods,the proposed scheme is motivated by image source identification.By detecting the influence of the fake(morphed)face generation process on the camera sensor pattern noise,face morphing spoofing attack can be detected.Firstly,a camera sensor pattern noise extraction method based on guided image estimation is proposed,which can effectively suppress the interference of image scene content.Then,a novel quantification statistics feature extraction method is used for effective identity the real face and fake(morphed)face.Thirdly,a face de-morphing generative adversarial network is proposed to restore the face morphing accomplice’s facial image.Different from the previous method,which needs prior knowledge of the morphing parameters in face morphing process,the proposed network restores the accomplice’s facial image without prior knowledge of the morphed facial images.It utilizes symmetric dual network architecture and two levels of restoration losses to separate the identity feature of the morphing accomplice.To the best of our knowledge,it is the first attempt to exploit a learning-based generation approach for facial image restoration in face morphing detection.Experimental results and analysis demonstrate the effectiveness of the proposed scheme.It has great potential to be implemented for detecting the face morphing accomplice in a real identity verification scenario.Fourthly,a benchmark database HNU-FM is built for providing a public and fair comparison platform for evaluating the performance of face morphing detection.The existing researches on face morphing attack detection have some problems,such as unbalanced sample sizes of real face images and morphed face images,fixed morphing factors,and no publicly available face morphing database.Therefore,a benchmark database HNU-FM is built in this paper.Four evaluation protocols are developed for different scenarios.The sample sizes in each protocol are sufficient and balanced.Furthermore,with HNU-FM,performance comparison is made to some existing face morphing detection methods,and their reliability with morphing factor variation is also analyzed.It has great potential to be used as a public evaluation benchmark database for face morphing detection.