Spatial LSB (least significant bit) matching was modeled as adding independent noise to the image, and its influence on the image histogram and the correlation between the adjacent pixels were analyzed. Accordingly, the absolute differences between adjacent elements of image histogram were calculated as the histogram features, and co-occurrence matrix was utilized to extract features based on image correlation. A calibrated image was generated by embedding message into the test image, and the features were extracted from both the test and calibrated images. The ratios of corresponding features between test and calibrated images were used as the final features. Support vector machines were utilized to train and test the classifiers on a JPEG (joint photographic experts group) compressed and an uncompressed image databases. The results show that which that exploit the histogram disturbance are better in terms of detecting uncompressed images, while the features based on the image dependence are accomplished for the detection of images with low noise. The proposed method utilizes both the kinds of disturbance and thus performs well.%将空域LSB(least significant bit)匹配嵌入模拟成像图像中添加独立噪声,分析LSB匹配嵌入对图像直方图和图像相邻像素之间的相关性的影响,计算图像直方图相邻元素绝对差作为直方图特征,运用共生矩阵模型对差分图像进行统计以提取图像相关性的特征;将检测图像嵌入信息构造1幅对应的校准图像,分别从待检测图像和校准图像提取特征,将对应特征的比值作为最终特征组成特征向量.在JPEG(joint photographic experts group)压缩和未压缩的2个图像库上利用支持向量机对特征向量进行训练和测试,并与已有算法进行比较分析.研究结果表明:基于图像直方图的特征在检测未压缩的图像时更具优势,而基于图像相关性的特征则更擅长检测含噪声较少的图像里的隐藏信息.该算法全面考虑了LSB匹配对图像直方图和图像相关性的影响,并用校准图像对特征进行校准,因而获得了良好的检测效果.
展开▼
