Text Steganography Based On Statistical And Semantic Distortion Model

The use of text data is increasing day by day with the rapid development of internet technology.This provides a natural environment for text steganography.The application prospects of text steganography is positive.Therefore,it is significant to design a secure and fast steganography algorithm that is suitable for text cover.Based on the existing text steganography technology,we develop a content adaptive steganography algorithm that is suitable for text cover.The main research contents and innovations consists of the following three points:1)Distortion quantification based on synonym frequency and word vector:For the problem that the synonym substitution will cause statistical distortion and semantic distortion,this thesis proposes a possible distortion quantization scheme.Use word frequency information to quantify the statistical distortion.Use the word vector model to train the word vectors and quantify semantic distortion according to the vec-tor distance.A variety of distortion quantification formulas are designed based on this scheme.The one of quantification formulas which is most practical is decided through experiments.Experiments show that the distortion quantification scheme is reasonable.2)Text steganography based on double-layered STCs embedding:For the problem that mainstream text steganography based on the synonym substitution can not resist statistical and semantic attacks,this thesis proposed a new text steganogra-phy.Use a double-layered STCs embedding method to minimize the distortion caused by synonym substitution.Experiments show that the proposed text steganography can resist statistical and semantic attacks better compared with the traditional text steganography based on synonym substitution.3)Fold embedding:For the problem that text steganography based on double-layered STCs is time-consuming to embed in large texts even if the embedding rate is low,this thesis designs a method to speed up the message embedding process while maintaining the security of the embedding method.Fold the cover to construct a half-length cover.Redefine distortion for the half-length cover.Experiments show that fold embedding can speed up the message embedding process without affecting the security performance.