Research On High-undetectability Audio Steganography Scheme Under STC Steganography Framework

Steganography,as an essential technique for information security,can hide data into various multimedia covers and guarantee the undetectability and security of data.In cyberspace,with the development of multimedia social platforms,the endless number of digital music and speech will become ideal potential covers for steganography.However,most of the existing audio steganography methods are facing many challenges,in theory,method,and applications.The most critical issue is how to improve the undetectability of the message under a large steganographic message capacity.This thesis aims at improving and optimizing the undetectability of the original STC(Syndrome-Trellis Codes)steganography framework,under a large message capacity.There are four main research contents: constructing a comprehensive audio database,improving the STC steganographic algorithm,enhancing the complex cover selection,and optimizing steganographic traces.Hopefully,the results of this research can provide new insights and practical guidelines for audio steganography and information security.The research contents of the thesis are summarized as follows.First,the current research status of the STC steganographic framework are reviewed.The basic researching ideas of the STC steganographic framework are summarized in detail.The rules,characteristics,and several issues of the current adaptive steganography methods are surveyed indepth.According to research contents,specific indicators for evaluating the performance of proposed methods are determined.Three different types of audio databases,which are suitable for the following methods,are established for evaluation.Second,by optimizing the STC algorithm,we propose a new steganographic algorithm,which reduces the number of distortion elements significantly.Since distortion elements can further reduce w.r.t.steganographic algorithm improvement,we propose four rules for selecting the optimal submatrix,according to the four features of the submatrix.Next,the optimized Segment-STC(SSTC)algorithm is proposed,based on the fact that segmentation is beneficial for further reducing the number of distortion elements.The experimental results illustrate that under the same message payloads,the number of distortion elements produced by S-STC is lower than STC's,and the optimization rate of the number of distortion elements can reach over to 20% maximum.Third,we devise an appropriate audio complexity indicator,and then design a new distortion function to select complex cover areas adaptively.Researches have shown that the overall cost is lower when the chosen steganographic cover is more complex.Therefore,an audio complexity indicator is designed.The original cover is convoluted with a specially designed adaptive convolution kernel.Based on the residual between the original and the convoluted audio,we derive a quantity for measuring the complexity of each frame for an audio clip.Then combined with the complexity indicator,we propose a distortion function,which is suitable for multiple LSB(Least Significant Bit)levels.The overall distortion cost can be accurately calculated and controlled by this function.The experimental results show that the complexity indicator can effectively filter the complex frames from different types of audio.The proposed distortion function can balance the perceived audio quality and the undetectability of the message more perfectly under the sizeable message capacity.Fourth,by studying the steganographic traces of the audio stego,we propose a post-processing method for optimizing steganographic traces,and then we propose the high-undetectability audio steganography method.Some studies have proved that with fewer steganographic traces,the undetectability of the message can be higher.Therefore,we propose the JSD-Based Trace Optimization Model(Jensen-Shannon Divergence,JSD)to optimize the stego audio without affecting the embedded message extraction,so that the distribution of the stego can be closer to the original cover one.Meanwhile,the modified positions will be controlled by the SNR(Signal-toNoise Ratio)threshold adaptively.The experimental results illustrate that the proposed postprocessing model is much effective than the other models,on the undetectability experiments.Ultimately,combined with the previous research contents of steganographic algorithm and cover selection,the final high-undetectability audio steganography method is proposed,which can achieve large message capacity.Experimental results prove that the undetectability of the message has been significantly improved.On different types of databases,it has increased by 4% to 21% against traditional feature-based steganalysis methods and 10% to 20% against CNN-based steganalysis methods.The optimization effect improves w.r.t.the message capacity expands.