Image steganalysis method based on adversarial training and key path extraction

Abstract

The invention discloses an image steganalysis method based on adversarial training and critical path extraction, and the method comprises the steps: respectively inputting a to-be-detected image intoa steganalysis module after adversarial training and a steganalysis module based on critical path extraction, wherein the to-be-detected image comprises a non-steganography image, a traditional steganography image and an adversarial embedded steganography image, and the traditional steganography image and the adversarial embedded steganography image belong to the category of the steganography image; obtaining the probability that the to-be-detected images output by the two modules are steganography images, and obtaining the probability that the final to-be-detected images are steganography images by combining the weights of the two modules and adopting a weighted fusion mode. According to the method, the steganography image generated through adversarial embedding can be detected, and meanwhile, the problem that the accuracy of a traditional steganography image and a non-steganography image is reduced due to adversarial training is balanced to a certain extent.

Description

technical field [0001] The invention relates to the technical field of image steganalysis, in particular to an image steganalysis method based on adversarial training and key path extraction. Background technique [0002] As a common means of information hiding, digital image steganography not only provides people with a more covert communication method, but also provides a more reliable link for individuals and organizations with improper intentions to carry out a series of activities that endanger national and social security. Way. [0003] In recent years, with the successful application of artificial neural networks in image classification and other tasks, many scholars have introduced convolutional neural networks into the field of steganalysis, and achieved good detection results in mainstream steganographic methods. In this context, in order to improve the anti-analysis of steganographic methods, some works have begun to try to apply the method of generative adversar...

AI Technical Summary

Problems solved by technology

This kind of steganographic image that has undergone confrontational embedding is difficult to be detected by the existing analysis network. At present, scholars have proposed to use confrontation training to solve this problem, but confrontation training also makes the classification of traditional steganographic images and non-steganographic images difficult. Accuracy drops

Images