Research On Image Steganography Based On Texture Analysis

Information hiding is regarded as an important means to ensure data security.Image is the most common kind of data media,which is often widely spread on the network and has certain security risks.At the same time,there are some special application scenarios,such as the need to embed URL and other information in the image,which makes the image steganography technology more concerned by researchers.Some studies have discussed the steganography capacity and robustness,but the quality of the generated steganographic image is unstable,some steganographic images have obvious change traces in the low-frequency region,and the processed image is fuzzy and of low quality,which is far from the cover image.This thesis takes the generation of high-quality steganographic image as the research object.Its purpose is to make the image have no obvious fuzzy phenomenon after adding a certain scale of information,which is not easy to be detected by the naked eye,and ensure to embed more information and maintain the robustness of the algorithm as much as possible.Therefore,this thesis proposes a based on attention mechanism and generative adversarial network model of image steganography using encoding-decoding network realization of information embedding and extraction,introducing the attention mechanism of learning important image features,introducing discriminator to train against the encoder,at the same time,an interference layer is added between the encoder and decoder to learn the disturbance in the process of image transmission and ensure the robustness of the steganography model.In order to further improve the quality of the steganographic image,the image steganography method based on texture analysis is adopted,the texture information is extracted by using the gray level co-occurrence matrix,and the information is embedded in the high complexity area of the cover image using the image steganography model based on attention mechanism and generation adversarial network.The experimental results using the public data sets MIRFLICKR and VOC2012 show that compared with similar methods,when embedding the same amount of information,the steganographic image obtained by the proposed image steganography model has higher PSNR value and SSIM value,and has good robustness.Especially after combining based on texture analysis,the PSNR value and SSIM value of the steganographic image have been greatly improved,and there is basically no change trace on the generated steganographic image.Compared with the non selective embedding information of other methods,the secret information of this algorithm is more embedded in the region with high image complexity,which has a higher degree of coincidence with the contour of the cover image,and the visual quality of the generated steganographic image has been significantly improved.At the same time,when the amount of embedded information increases by about 50%,the processed image can still maintain the level of similar algorithms,which makes this algorithm can embed more information under the same quality requirements.