Method and system for detecting camouflage attack resistance of face recognition system

Abstract

The invention discloses a method and a system for detecting camouflage attack resistance of a face recognition system. The method comprises the following steps that a face image X and a face image Y are acquired; wherein the face image X corresponds to a classification number M, and the face image Y corresponds to a classification number N; noise Z is added to the face image X by using an iterative attack method based on gradient and momentum, the noise Z is masked into the shape of a wearable article, and a face image G containing noise is obtained; the face image G attacks the to-be-detectedface recognition system, and if the output classification number is N, it is indicated that the to-be-detected face recognition system fails to resist the directional camouflage attack; and if the output classification number is M, it is indicated that the to-be-detected face recognition system successfully resists the camouflage attack. The face recognition system is attacked by using the generated wearable decorative article with the camouflage characteristic, whether the face recognition system has the capacity of resisting the camouflage countermeasure sample can be detected, and the safety of the face recognition system is evaluated.

Description

technical field [0001] The invention relates to the field of computer vision and the field of deep learning technology, in particular to a method and system for detecting a face recognition system against masquerading attacks. Background technique [0002] With the rapid development of computer vision and artificial intelligence technology, face recognition technology has become more and more widely used, such as face recognition access control attendance, electronic passports and ID cards, self-service entry at high-speed rail stations, etc. However, with the development of technology, the risk of face recognition system being attacked by spoofing is increasing day by day. For example, in the face recognition system widely used in video surveillance and access control, if some adversarial samples are input into the face recognition system, the face recognition system will be misled by the adversarial samples, and the face recognition will be wrong, thereby threatening perso...

AI Technical Summary

Problems solved by technology

However, the existing adversarial sample generation method for Rosenbrock-PSO is only used to find the adversarial samples that make the face recognition system go wrong, and it cannot guarantee that the face recognition system can resist the attack of directed camouflaged adversarial samples, that is, the adversarial samples generated by this method The samples conduct confrontational training on the face recognition system, and any face recognition system obtained after training is still unable to detect the ability to resist the adversarial sample attack generated by the face image disguised for a specific target

Moreover, the face recognition attack defense method based on Rosenbrock-PSO is only a way of adding confusing data to the training set. By interfering with the training process of the machine learning model, the machine learning model has more wrong predictions, but this method has the defect of low applicability , the attacker generally cannot obtain and interfere with the source of the training set of the target model algorithm, and is generally only applicable to the experimental environment

Images