Research on Image Multi-Feature Steganography Based on Deep Learning

Digital steganography is one of the important branches of information security. It hides information in common transmission carriers in life, such as digital images, digital video, digital audio signals, etc., hiding the existence of information. At present, digital image steganography has become one of the research hotspots in the field of information security. Steganalysis technology uses the change of information embedding to the statistical characteristics of the original carrier to determine the existence of hidden information, so as to detect whether information is embedded in the carrier. In this paper, based on DL (Deep Learning), the steganographic analysis of image multiple features is further studied. The image multiple features can be divided into gray image, color image and binary image according to the difference of color and value bits represented by them. Grayscale images only express the brightness value of image pixels, but not the color information of image pixels. In view of the particularity of image steganography analysis task, the general DL network is specially designed, and the accuracy and speed of analysis are studied by changing the network structure, network layers, activation function and adding filter layer convolution.