Adversarial batch steganography method

Abstract

The invention discloses an adversarial batch steganography method, which introduces an adversarial sample (namely an image of a normal user) concept, uses a loss function to adjust a distortion function about the gradient of the image, and improves steganography security. The anti-secret-carrying graph obtained on the basis of the method is superior to an existing method in resistance to pooling steganalysis based on deep learning; meanwhile, the method can also resist steganalysis of a single image to a certain extent.

Description

technical field [0001] The invention relates to the technical field of adversarial samples and steganography, in particular to an adversarial batch steganography method. Background technique [0002] Steganography is a covert communication technique that conceals messages within a carrier (such as a digital image) without arousing suspicion. Based on the STC steganography framework, the current steganography has matured, such as HILL, UNIWARD, UERD, etc. These algorithms are all based on the laboratory environment, but in practical applications, steganography generally uses multiple images as carriers, so some scholars propose a batch steganography strategy. [0003] Batch steganography takes a group of images as a unit, and studies the message distribution strategy among images based on the steganography algorithm of a single image. Ker et al first proposed four batch steganography strategies in 2012, namely, average strategy, random strategy, maximum greedy strategy, and...

AI Technical Summary

Problems solved by technology

Since pooling steganalysis usually uses unsupervised learning methods, single-image adversarial steganography algorithms cannot be directly applied to batch steganography

Moreover, there is not a directly derivable loss function in pooled steganalysis that can be directly used for steganography

Images