Research On Universal Steganalysis Technology Based On Machine Learning

With the popularity of the network and the rapid development of multimedia communication technology, the way people transmit information is more and more diversified. At the same time, the widespread use of the Internet has resulted in many security problems, Information hiding technology is receiving more and more attention. Steganography and steganalysis are two important research topics in the field of information hiding. In recent years, steganalysis has been developing rapidly, but steganography is still ahead of steganalysis in general.In this paper, we study the general steganalysis problem based on machine learning and improve the generalization performance of the existing steganalysis algorithms, and propose a steganalysis algorithm based on convolutional neural network. The main work is as follows:(1)Summarize the principles and the evaluation criteria of steganography and steganalysis. Several typical steganographic algorithms and steganalysis algorithms are studied in detail. The basic principles and algorithm steps of the algorithms are introduced. And several commonly used classifiers are introduced.(2)An improved steganalysis algorithm based on training set construction is proposed. The generalization performance of Rich and Model SPAM is analyzed by different experimental designs, including steganographic mismatch, steganographic mismatch and image source mismatch. Based on the experimental results, several schemes are proposed to solve the generalization problem by training a small steganographic image set, training a variety of steganographic algorithms, image preclassification and improved IQM analysis algorithm. The experimental results show that the performance of the improved algorithm is significantly improved.(3)A steganalysis algorithm based on convolutional neural network is proposed. Firstly, the structure of artificial neural network is introduced. Then the structure of convolutional neural network and the AlexNet network model are introduced. And combined with the advanced deep learning framework, the AlexNet network is fine tuned to classify the watermark images and the original images. Experimental results show that the proposed method is better than SPAM but worse than Rich Model.