An image steganography method based on texture perception

Existing image steganography technology can resist image attacks, however, the quality of generated encoded image is poor and the low-frequency region can be perceived. To improve the quality of the encoded image, an image steganography method based on texture analysis is proposed. In this method, the gray-level co-occurrence matrix is used to extract the texture information of the cover image, the information is embedded in the complex position of the image, in and the network structure of the StegaStamp encoder is improved, small convolution layers are added in the down-sampling process to expand the receptive field, and channel attention is used to assign different weights to feature channels to enhance the influence of important features of images. The experimental results show that the proposed method can effectively improve the PSNR and SSIM values of coded image, and the quality is improved without degrading the robustness and embedding capacity.