Adversarial sample generation method and device, storage medium and electronic equipment

Abstract

The invention relates to an adversarial sample generation method and device, a storage medium and electronic equipment, and aims to improve aggressiveness of generated adversarial samples so as to defend adversarial attacks more effectively. The method comprises the following steps: acquiring an original image of a to-be-generated adversarial sample; determining a target standard deviation image with the same size as the original image, and determining a target Gaussian blurring kernel corresponding to each pixel point in the original image according to the target standard deviation image and a preset blurring radius; for each pixel point in the original image, determining a target Gaussian blur value corresponding to the pixel point according to the target Gaussian blur kernel and target pixel points located around the pixel point and within the preset blur radius range, so as to obtain a target Gaussian blur image corresponding to the original image; and taking the target Gaussian blurred image as an adversarial sample corresponding to the original image.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, and in particular, to a method, device, storage medium, and electronic device for generating an adversarial example. Background technique [0002] With the large-scale application of various image processing models, attacks against image processing models emerge in an endless stream. It is necessary to follow up research in time to discover potential attack methods and prevent dangers before they happen. Among many attack methods, adversarial attack is a new type of attack method with strong aggressiveness. Adversarial attacks cause image processing models to give a wrong output with high confidence through adversarial examples. Among them, adversarial samples are constructed by attackers through specific means. Human eyes look no different from normal samples, but a type of image that the machine will misclassify. [0003] In related technologies, the generation of adversaria...

AI Technical Summary

Problems solved by technology

[0003] In related technologies, the generation of adversarial samples is usually based on applying pixel-level perturbations to the image, which can be regarded as imposing some special noise on the image. Therefore, such adversarial samples can be generated by introducing a feature-level denoising module. To a certain extent, the attack is not strong, so the image processing model is trained through this type of adversarial samples, and the defense capability of the trained image processing model is also lacking, and it is impossible to achieve effective defense against other types of adversarial samples.

Images