Adversarial sample generation method of face recognition model and related device

Abstract

The invention discloses an adversarial sample generation method for a face recognition model, and the method comprises the steps: carrying out the decoupling of each feature of a face image based on a hidden layer of the face recognition model, and obtaining a plurality of features; performing result influence score calculation on each feature based on preset disturbance data to obtain an influence score corresponding to each feature; and adding the features of which the influence scores are greater than a threshold value to the face image to obtain an adversarial sample. Important features in a face image are screened out, then the important features replace an original image to serve as model input, and an adversarial sample is generated, so that unnecessary disturbance is reduced, disturbance is limited in a small area range, the situation that an attack image is excessively distorted due to large disturbance is avoided, and the effectiveness of the adversarial sample is improved; and thus, the safety of training the face recognition model is improved. The invention also discloses an adversarial sample generation device of the face recognition model, a server and a computer readable storage medium, which have the above beneficial effects.

Description

technical field [0001] The present application relates to the technical field of machine learning, and in particular to a method for generating an adversarial example for a face recognition model, an adversarial example generation device, a server, and a computer-readable storage medium. Background technique [0002] In the field of artificial intelligence, face recognition technology is based on deep neural network to extract face features. With the development of technology, the deep neural network is very vulnerable to the attack of "adversarial samples". By adding small perturbations that cannot be detected by the human eye, the model can make incorrect predictions. Common security risks include escape recognition, that is, the Dangerous elements are misidentified as any other face, bypassing the security check of the face recognition system. Or the person is misjudged as another specific person to be recognized by the system as a specific authorized user, so that the i...

AI Technical Summary

Problems solved by technology

However, the validity of the adversarial samples obtained in related technologies is low, and there is no way to improve the security of the model.

Images