Image Steganalysis Algorithm Based On Deep Learning

As a hot spot in the information security area,steganography and steganalysis technologies have been making progress together in confrontation with each other.Once steganography is used illegally,it will threaten social stability and even national security.Therefore,the development of steganalysis technology is very necessary.To promote the development of steganalysis technology based on deep learning and reduce the use of prior knowledge and manual participation in steganalysis technology,an end-to-end deep learning capable of steganalysis of images with different resolutions is designed.The main work of the steganalysis algorithm is as follows:First,an image steganalysis model based on deep learning is designed.In the preprocessing part,an image denoising module based on convolutional neural network is trained to obtain clean and noise-free image content,and then the steganographic noise residual of the image can be obtained by subtracting this clean part from the input image.Retain as much steganographic noise information as possible while reducing human involvement.The rest of the model is obtained by improving Yedroudj-Net.To improve the network effect,the convolutional layers of the first two layers are replaced by depthwise separable convolutions to obtain the spatial correlation of features,and at the same time,the network is reduced.The amount of parameters speeds up the calculation of the network.Not only that,but this paper also deepens the network,so that the model can extract more features,thereby improving the detection effect.In addition,in order to detect images of different scales without causing information loss,spatial pyramid pooling is added between the convolutional layer and the fully connected layer.Second,in order to verify the performance of the designed model,eight experimental required datasets are established.First crop the images in the BOSSBase1.01 dataset into images of two resolutions.Then use WOW steganography technology to embed images of each resolution with 0.2bpp and 0.4bpp embedding rates to obtain two secret data sets,and then use S-UNIWARD steganography algorithm to embed 0.2bpp and 0.4bpp The rate-to-image embedding of secret information yields two other secret-bearing datasets.In this way,four density datasets can be generated for each resolution image,and a total of eight density datasets are generated for the two resolutions.Finally,use the model designed in this paper to conduct steganalysis on the data set constructed in this paper,and use the accuracy rate as a standard to measure the detection effect of the model.Experiments show that the model realizes steganalysis of images,the designed image denoising module for preprocessing can effectively extract noise residuals in images,and the spatial pyramid pooling is used to realize the model’s ability to detect different resolutions.Steganalysis of images.