压缩域语音频隐写的失真函数构造方法研究

With the rapid growth of sharing audio/video data over social networks, digital speech-audio steganography recently is becoming a very high-profile research field. However, due to the frame-offset effect (FOE) of audio encoders, and the primary theory of constructing distortion function for audios, research on the adaptive steganography for compressed speech-audio faces severe challenges. In this project, we will research on the adaptive steganography framework for compressed speech-audio and the universal distortion function theory, which provides a complete theoretical basis and technological reserve. First, through analyzing the mechanism of audio codecs and the structure of codestream, a meta embedding model of code-structure preserving and a frame-level perception distortion (FPD) model are built for designing a general multi-layer adaptive audio steganography framework, which simultaneously achieves both adaptive steganographic embedding for the intra-frame and the inter-frame. Second, based on the mathematical model of auditory masking effect of psycho-acoustical theory and the steganalysis mechanism of Markov model, we respectively research measurement methods of the audio content distortion and the statistical distortion. Lastly, integrating content distortion and statistical distortion model based on non-additive distortion mechanism of multi-nary steganographic coding is established for researching a series of cost function methods according to joint the auditory perceptual distortion and the statistical feature preserving, which is able to improve the anti-steganalysis capacity of adaptive steganographic algorithms.

随着社交网络中音视频分享内容的迅猛增长，针对数字语音频载体的信息隐写已成为研究的热点。然而受到音频编码“帧偏移效应”制约，以及适用于音频内容的失真代价函数构造理论不完备等因素的影响，压缩语音频自适应隐写研究面临严峻挑战。本项目拟对适用于压缩语音频的自适应隐写框架及通用的失真函数理论开展研究，为音频自适应隐写方法提供较完备的理论基础和技术储备。首先，通过分析音频编码器的机理和压缩编码结构，建立码字结构保持的元嵌入操作模型和帧级感知失真度量的数学模型，研究一般化的多层自适应隐写框架，同时实现帧内帧间自适应隐写嵌入；其次，利用基于心理声学理论的听觉掩蔽效应数学模型和基于马尔可夫模型的隐写检测机理，分别研究音频内容失真和统计失真的度量方法；最后，基于多元隐写编码的非加性失真机理构建融合内容失真和统计失真的数学模型，研究联合听觉感知失真和统计特征保持的代价构造方法，提高自适应隐写算法的抗分析能力。

随着语音和音频社交网络的流行，面向数字语音频载体的自适应隐写研究受到重点关注。然而受到音频编码“帧偏移效应”制约，以及适用于音频内容的失真代价函数构造理论不完备等因素的影响，压缩语音频自适应隐写研究面临严峻挑战，因此本项目对适用于压缩语音频的自适应隐写框架和通用的失真函数构造方法开展研究，形成的主要研究成果包括：（1）提出了MP3音频Linbits域的失真代价函数构造方法，（2）提出了AAC音频熵编码域的自适应隐写方法，（3）提出了基于音频编码联合域嵌入的快速隐写方法，（4）提出了转码鲁棒的MP3隐写方法，（5）提出了基于隐藏信息内容感知的鲁棒隐写方法，（6）提出了一系列音频隐写分析方法。项目提出的一系列压缩域自适应音频隐写算法解决了MP3和AAC等主流音频编码域的自适应隐写框架和通用的失真函数构造等科学问题，极大地充实了自适应音频隐写理论与方法，也推动了音频隐写技术走向成熟及应用。