Abstract: It is hard to detect the ±1 steganography when digital media are scanned images.This paper proposes a novel targeted steganalysis method for this issue.We found the fact that after ±1 embedding，the local maxima of an image histogram decrease and the local minima increase.Furthermore，based on an additive noise model for ±1 steganography，it is similar in wavelet histogram.That means local extrema of histograms are sensitive to ±1 embedding method.As a result，a 5-dimensional feature vector is constructed for detecting stego images.Experimental results show the detection performance of proposed method is superior to some methods in the literature.

Key words: embedding, steganalysis, local extrema, histogram

摘要： 当媒体为扫描图像时，对±1嵌入方法的隐秘分析较为困难。为此，提出了一种新的直接隐秘分析方法。在用±1方法嵌入秘密信息以后，图像直方图的局部极大值变小而局部极小值变大。进而，基于±1嵌入方法的加性噪声模型，小波直方图也有相似的结果。这意味着图像直方图及其小波直方图的局部极值对嵌入方法比较敏感。据此，构造了一个5维特征向量来检测隐秘图像。实验结果表明，该方法的检测性能优于相关文献方法。

关键词: ±1嵌入, 隐秘分析, 局部极值, 直方图