Research On Image Steganography Based On Embedding Domain Selection

Steganography is a technique of covert communications by hiding the secret information into innocuous cover media. Its countermeasure, steganalysis, aims to to attack the steganography by detecting the presence of the secret information, and hence to deter covert communications. Battles between steganography and steganalysis have become an important issue in information hiding, which makes steganography and steganalysis restricts mutually, and promote mutually. This paper is focused on the techniques of steganography and steganalysis in images. The main content of this thesis is summarized as follows:1. Improved existing steganographic methods by embedding data in a parametrized wavelet domain instead of a fixed one, and proposes the key-dependent wavelet (KW) image steganographic method which uses the parameters as its secret key. Since the embedded signal disperses in the domain guessed by an attacker, the KW method makes steganalysis difficult to detect the characteristics of steganography. The analysis and experiments show that it improves the covertness of embedded signal by becoming more resistant to the steganalysis based on testing the commonly used test statistics including pairs of value, gradient energy, histogram, wavelet characteristics, etc.2. Steganography based on quantization index modulation (QIM) is being increasingly threatened by steganalysis. The paper improves the existing steganographic methods by embedding data in non-uniform discrete cosine transform (NDCT) domain instead of more commonly used discrete cosine transform (DCT) domain, and proposes the NDCT-QIM image steganography which uses the parameters of NDCT as its secret key. Since the embedded signal disperses in the domain guessed by an attacker, the NDCT-QIM method makes steganalysis difficult to detect the characteristics of steganography. The analysis and experiments show that it improves the covertness of embedded signal, and is more resistant to the commonly used steganalysis methods, including those respectively based on testing gradient energy, histogram, wavelet characteristics, etc.3. Based on non-uniform discrete cosine transform (NDCT), we propose an adaptive domain selection steganography (ADS Steganography) by embedding data in a selective domain with minimum distortion. Since the distortion in domains which the steganalysis may use decrease, ADS Steganography can resist steganalysis better. The experiments show that the detection rates of some recent blind steganalysis schemes are low.4. To measure the security of our three proposed steganography methods:KW, NDCT-QIM, ADS, we use three classic universal steganalytic scheme:steganalysis based on higher-order statistics, steganalysis based on histogram features, steganalysis based on multiple features formed by statistical moments of wavelet characteristic function to detect, and compare to some steganography methods: Jsteg, OutGuess, JPHS, F5.