Image steganalysis method with unknown embedding rate

Abstract

The invention provides an unknown-embedding-rate image steganalysis method, which comprises the following steps of: constructing a pre-training model based on a first training set accommodating a carrier image and a high-embedding-rate secret-carrying image; on the basis of the pre-training model, obtaining a pre-training model; a classification task used for distinguishing a carrier image and a secret-carrying image is used as a main task; a regression task used for fitting the embedding degree of the secret information in the image is added to serve as an auxiliary task so as to construct asteganalysis model based on multi-task learning, and the parameter initial value of the steganalysis model based on multi-task learning migrates from the parameter value of the pre-training model; anddesigning a target function based on the multi-task learning steganalysis model to perform steganalysis. According to the technical scheme, the performance of model steganalysis blind detection can be effectively improved, and meanwhile the defects of a manual weight adjustment method are overcome.

Description

technical field [0001] The invention relates to the technical field of information hiding, in particular to an image steganalysis method with an unknown embedding rate. Background technique [0002] Image steganography uses the spatial redundancy of images to hide a meaningful secret information in the carrier image, thus obtaining a secret image. Image steganalysis detects whether an image contains secret information by distinguishing carrier or secret images, which is usually regarded as a binary classification problem. Image steganalysis methods can be divided into specific steganalysis detection and general blind detection. Specific steganographic detection is to judge whether the image contains hidden information when the steganographic algorithm and embedding rate of the image are known. Universal blind detection is to judge whether an image contains hidden information when the steganographic method or embedding rate is unknown. Although the detection accuracy of ge...

AI Technical Summary

Problems solved by technology

The existing general blind detection method of steganalysis based on deep learning still needs information such as steganographic algorithm or embedding rate, and cannot truly achieve blind detection

There is a steganalysis model based on multi-task learning proposed in related literature. The weight distribution involved needs to be adjusted manually. The model must be retrained every time the weight is adjusted, resulting in a large consumption of computing resources and time. Therefore, the analysis Blind detection performance is low

Images