A generative information hiding method combining difference clustering and minimum error texture synthesis

Abstract

The invention provides a generative information hiding method combining difference clustering and minimum error texture synthesis. Firstly, the clustering center is obtained through the difference mean clustering of a sample block and a random key template, and the multiple mapping relationship between the secret information segmentation and the cluster center sample block is established. Then, combined with the random placement position, the appropriate sample block is selected from the sample block clustering center as the encoded representation of the secret information segmentation to encode the secret information and place it on the designated position of the blank image. Finally, the closest sample block is selected from all the sample blocks according to the minimum error texture synthesis strategy to conceal the secret information and generate texture image. Compared with the existing methods, the proposed method can effectively conceal the secret information by generating high-quality texture images, and the secret information segmentation does not correspond to the sample blocks in the clustering center, and the recovery process is completely dependent on the key and hasa strong resilience to attacks.

Description

technical field [0001] The invention belongs to the cross field of image information security and digital image signal processing, relates to an information hiding method, in particular to a generative information hiding method combining difference clustering and minimum error texture synthesis. Background technique [0002] With the continuous deepening of deep learning, the continued development of artificial intelligence and the emergence of quantum computers, the traditional multimedia information security situation with image and audio as the main transmission medium has become more severe. On the one hand, the dimensionality of steganographic classifiers based on machine learning continues to increase, making it less and less likely that traditional modified information hiding hidden information will not be discovered; on the other hand, the continuous development of compression technology also makes traditional modified The redundant space available for information hi...

AI Technical Summary

Problems solved by technology

[0006] However, the problems with this type of method are: ① normal text and images have very limited ability to express irrelevant secret information, if the text and images in the database need to have sufficient expression ability, it is necessary to collect enough samples; With the enhancement of database text and image information expression capabilities, the amount of data to be searched is also increasing geometrically. Even with the help of inverted index, it will be a heavy burden for the search, storage and maintenance of large amounts of data, but with this At the same time, the embedding capacity is extremely low, involving intensive transmission of a large number of text and image collections in public channels; ③Although the isolated text or images of this method are normal text or images, which cannot detect abnormalities because they have not been modified, a large amount of text and images The dense transfer of image collections is also bound to raise suspicions

[0012] ①The same sample block usually represents the same secret information at different positions, so a specific sample small block is directly associated with a specific secret information segment, resulting in the leakage of secret information;

[0013] ②Although some methods introduce classification, one-to-one encoding is performed on the secret information segment by the category of the sample block, which leads to a fixed one-to-one relationship between the secret information segment and the sample block category. Correspondence, low security;

[0014] ③ Divide the small sample blocks into different categories, randomly select the small sample blocks from the same category, without considering the difference between the small sample blocks themselves and the sample small blocks that deviate from the classification center may lead to category extraction errors when attacked, Therefore, the ability to resist attacks is very limited

However, the error bar starting from a specific starting position does not mean the minimum error bar of all the smallest difference pixels in the overlapping area. At the same time, when there are multiple adjacent overlapping blocks, if the splicing order of non-adjacent block overlapping areas is not appropriate, It will be easy to produce traces of splicing, so that secret information cannot be concealed

Images