Image forensic adversarial attack defense method based on class feature restoration fusion

Abstract

The invention discloses an image forensic adversarial attack defense method based on class feature restoration fusion, and the method comprises the steps of inputting an original image containing adversarial disturbance into a classification model constructed based on a convolutional neural network, carrying out the class feature visualization of an extracted feature map, obtaining a reconstruction region based on a visualization result, and obtaining an adversarial disturbance model; performing feature restoration on the reconstructed area to obtain a restored image; and denoising the original image to obtain a denoised image, fusing the restored image and the denoised image to obtain a fused image, selecting a plurality of classification areas with high image quality according to the pixel distribution characteristics of the fused image, inputting the classification areas into the classification model, and taking the highest classification probability output by the classification model as a defended class label. The method can improve the robustness of the deep learning model and accurately restore the evidence obtaining information.

Description

technical field [0001] The invention relates to the field of data security, in particular to an image forensics adversarial attack defense method based on class feature restoration and fusion. Background technique [0002] With the rapid development of electronic technology and the rapid popularization of digital cameras and image scanning equipment, digital images have been widely used in people's daily office, study and life, and are widely used in key fields such as media publicity, military intelligence, judicial identification, and scientific discovery. application. In recent years, due to the superior performance and stable performance of the deep learning model, it has been applied in the field of image recognition and played an important role in image forensics. [0003] However, the deep learning model is vulnerable to adversarial perturbation and misjudgment. Adversarial attack algorithms in the field of computer vision, such as FGSM, DeepFool, BIM, JSMA, etc., c...

AI Technical Summary

Problems solved by technology

[0009] In order to overcome the shortcomings of existing image forensics methods that cannot restore the fingerprint information of the original image device and are difficult to defend against adversarial attacks, the present invention provides an image forensics adversarial attack defense method based on class feature repair and fusion, which can improve the deep learning model. Robust and accurate restoration of forensic information

Images