中文摘要：

隐写分析的研究一直集中于检测隐藏信息的存在性,而关于如何提取隐藏信息（即对隐写术的提取攻击）的研究还非常少.对于使用密钥的隐写术,提取攻击等价于恢复隐写密钥.文中结合隐写分析中的检测技术和密码分析中的相关攻击技术,对空域图像LSB隐写术提出了一种隐写密钥恢复方法.理论分析说明：此攻击方法的计算复杂度主要由所需的样本量决定,并且当嵌入率r接近0或1时攻击将失败.作者通过混合高斯模型给出了一个估计最小样本量的方法.针对隐写软件“Hide and Seek 4.1”的实验表明：此攻击方法可以成功恢复隐写密钥,从而提取隐藏的消息.如果消息长度L未知,当嵌入率5.3%

英文摘要：

Many steganalysis techniques have been developed to detect the existence of hidden information. However, there are few approaches about how to extract the hidden information, i. e. , extracting attack to steganography. For key-dependent steganography, extracting attack is equivalent to recovering the stego key. By combining detecting techniques in steganalysis and correlation attack in cryptanalysis, the authors propose a new method to recover the stego key of LSB steganography in spatial domain of images. Theoretic analysis is made for this method, which establishes that its complexity is mainly determined by the number of samples and the attack will fail when the embedding rate r tends to 0 or 1. The authors introduce an estimator for the minimum number of samples n based on a mixture Gaussian model. The experiments on the stego software ＂Hide and Seek 4.1＂ show that the proposed method can recover the stego key and extract the hidden message. If the length of messages L is not known, the attack can success for 5.3%〈r〈94.7%; if L is known, the attack can success for 1.1%〈r〈98. 4%, and by using the estimator for n, the attacking speed can be increased by 10%-45% when 11%〈r〈50%.