Adversarial image generation method, device and equipment, and readable storage medium

Abstract

The embodiment of the invention discloses an adversarial image generation method, device and equipment and a readable storage medium, and the method comprises the steps: generating a reference model equivalent to the classification of a target classification model according to the target classification model; obtaining a target image, and generating original noise for the target image according tothe reference model; inputting the first noise and the original noise into an adversarial model, and outputting second noise corresponding to the first noise when the adversarial model meets a convergence condition, wherein the second noise is the noise for enhancing the original noise information entropy; generating an enhanced noise image corresponding to the target image according to the second noise and the target image, wherein the classification accuracy of the enhanced noise image in the target classification model is smaller than that of the target image in the target classification model. By adopting the invention, the enhanced noise picture is difficult to identify by the malicious target classification model, and the security of the enhanced noise picture is improved.

Description

technical field [0001] The present application relates to the field of computer technology, and in particular to a method, device, device, and computer-readable storage medium for generating an adversarial image. Background technique [0002] In recent years, with the rapid progress of machine learning, classifiers based on Deep Neural Network (DNN) have become very important supporting technologies in various industries, including criminals using DNN classifiers to classify websites or applications. Malicious attacks, such as automatically identifying and unlocking pictures through DNN classifiers to perform some illegal operations, therefore, how to generate effective adversarial pictures to defend against illegal elements' target networks has become an important research direction. [0003] The noise sample characteristics in the adversarial pictures generated by the existing adversarial attack methods are too obvious. When the target network adopts corresponding defense ...

AI Technical Summary

Problems solved by technology

[0003] The noise sample characteristics in the adversarial pictures generated by the existing adversarial attack methods are too obvious. When the target network adopts corresponding defense methods, it is difficult for the adversarial attack to achieve the expected results. security against images

Images