Image steganography based on convolution neural network

Abstract

The invention belongs to the field of image processing and computer vision technology, and is a method capable of effectively extracting steganographic features from steganographic images and improving image steganographic analysis effect on the premise of restraining original image detail features. For this purpose, the technical scheme adopted by the invention is that, according to the invention, By training convolution neural network, minimize network losses, The optimal parameter value P* is obtained to improve the detection performance of the network model. After the image is processed bythe convolution neural network preprocessing layer and the feature extraction layer, the extracted features are classified by the full connection layer, and the extracted features are classified by the softmax layer in the full connection layer. The invention is mainly applied to image processing occasions.

Description

technical field [0001] The invention belongs to the technical fields of image processing and computer vision, and in particular relates to an image steganalysis method based on a convolutional neural network. Background technique [0002] People mostly regard the problem of steganalysis as a binary classification problem, and the goal is to distinguish normal images from encrypted images. The difficulty of image steganalysis is that the steganographic noise signal introduced into the image by steganographic operation is usually very weak, and the image difference before and after steganography is very small, and this difference is easily concealed by the difference between different image contents. And when the amount of information embedded in the image gradually decreases, the difficulty of steganalysis will further increase. In particular, the content-adaptive steganography proposed in recent years can preferentially hide the steganographic signal in complex texture area...

AI Technical Summary

Problems solved by technology

However, the design of effective features depends more on human experience and takes a lot of time

In the deep learning method, the convolutional structure in the model has advantages in capturing the correlation between pixels in the local area of ​​the image, but the convolutional neural network (CNN) utilizes global information in the process of feature extraction. Generally, the information of the local area is fused layer by layer through the scaling or pooling of the convolutional layer. This method has its own limitations.

Images