Adversarial face image generating method and device

Abstract

The invention discloses an adversarial face image generating method and device. The method relates to adversarial attacks in the field of computer security, and comprises the steps of obtaining a reference face image and an adversarial face image; through a pre-trained target face recognition model, extracting feature maps under multiple scales from the reference face image and the adversarial face image respectively; performing fusion processing on the feature maps of the reference face image under multiple scales to obtain feature information of the reference face image; performing fusion processing on the feature maps of the adversarial face image under multiple scales to obtain feature information of the adversarial face image; calculating a loss value of the feature information of thereference face image and the feature information of the adversarial face image; and based on the loss value, iteratively adjusting the pixel value of each pixel point in the adversarial face image toobtain a target adversarial face image. According to the invention, the versatility and mobility of the adversarial face image on the face recognition model can be improved by fusing the feature mapsof the adversarial face image under multiple scales.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a method and device for generating an adversarial face image. Background technique [0002] With the development of computer technology, image processing technology has been applied to more and more fields. For example, face recognition technology is widely used in many fields such as access control and attendance, information security, electronic certificates and monitoring security. Specifically, face recognition technology is a technology that automatically extracts face features from face images, and then performs identity verification based on these features. [0003] Usually, face recognition can be realized by extracting facial features based on deep neural network. However, deep neural network is vulnerable to confrontation attacks. Due to some defects naturally existing in deep neural network, such as sensitivity to noise, it is easy to be disturbed by noise ...

AI Technical Summary

Problems solved by technology

However, in the current related technologies, the generated adversarial samples have weak versatility and transferability on the face recognition model, and cannot be well used to test the face recognition model.

Images