将隐藏信息检测与图像内容分析相结合是当前提高图像隐写分析性能的一个新方向。与基于图像整体内容的检测方法不同，该文分析了最低有效位(Least Significant Bit LSB)匹配隐写对图像子区域统计特性的影响，提出一种新的联合判决检测方法。首先依据图像内容复杂度将整体图像分割为若干类子区域，其次采用两组不同的滤波器分类提取各子区域像素序列直方图频谱特征，之后用各类子区域特征分别训练Bayes分类器以获得其权重，最后对待测图像的每一个子区域进行分类检测，并将结果加权融合得到最终判决。实验结果表明，该方法对LSB匹配隐写的检测性能优于现有典型方法。
Recently, it is a new direction to improve the performance of image steganalysis by combining the detection of information hidden with image content analysis. Relative to methods depending on entire image, this paper analyzes the effect of LSB (Least Significant Bit) matching steganography on image sub-areas, and presents a novel steganalyzer based on the combined discrimination. Firstly, the images are divided into several sub-areas according to the image content complexity. Secondly, the histogram spectral features of pixel sequence of each sub-area are extracted by using two different filters. Then, the Bayes classifiers are trained respectively by features of each class of sub-area in order to obtain its weights. Finally, each sub-area of a test image is detected depending on its class and the final discrimination result of the test image is achieved by weighted fusion of the results of its sub-areas. Experimental results show that the proposed method exhibits excellent performance for the detection of LSB matching, outperforms existing representative approaches.