Spatial Image Steganography Algorithm Under Embedding Distortion Minimization

As an important branch of information hiding,steganography is a technique to hide secret messages in public multimedia data to achieve covert communications that will not to arouse suspicion,it has a special security that obtains much attention of scholars in the field of information security,and image steganography is the key research contents in the field.Design diversified,high security and high embedding rate is the key problem in the field of image steganography.In this paper,we will improve the cost function of existing steganographic algorithms under embedding distortion minimization theory.Secondly,we focus on utilizing deep learning to design steganographic algorithm.The main work and contributions of this paper are as follows:(1)A cost diffusion method proposed that utilizing maximum filter for steganographic cost to make the modificationn location more centralized.The current cost diffusion is mainly through the average filter,although it can make the cost of adjacent pixels closer,it is still not particularly ideal.This paper utilizes maximum filter to filter the cost,so the cost of adjacent pixels is very close,even equal.Experimental results show that the maximum filter can make information embedded in a more compact region,especially in rich texture,so can significantly improve the anti-steganalytic ability.It has a conclusion that the safety of modification of pixel decided by the local minimal security pixel.Combined with the conclusion,modified some cost function,and their security performance improved again.(2)A new method utilizing deep learning for steganography priority judgment proposed.The method utilizes convolution neural network to judge whether the pixel is suitable for steganography,completed in two steps.The first step is to establish a suitable training image database,simulation cover and stego images,utilizes the image database to train a classification of the convolutional neural network,it is actually a specific steganalysis network.The second step is to judge whether a pixel is suitable to embed message through make a same format image as the input of network to obtain the output probability,assign each pixel with suitable modification cost to embed message under embedding distortion minimization.This method utilizes steganalysis to steganography,the experimental results show that the proposed algorithm based on deep learning can automatically extract the texture features of image,and then judge the safety degree of each pixel in steganography,achieves the state-of-the art performance on resisting advanced steganalysis.