中文摘要：

针对图像隐写中，现有的特征提取方法难以有效反映图像的微小变化，提出一种基于分层次提取的图像局部纹理特征。先对图像高通滤波得到残差图像，将其分块并根据图像块的纹理复杂度对图像进行分割，对不同类型的子图像分别提取改进的LBP（局部二值模式）特征，将得到的高维特征通过Fisher判别进行降维，得到最终的特征。对空域LSBM、HUG0与S-UNIWARD算法在不同嵌入率下进行了实验，结果表明，该方法时间复杂度较高，但其隐写检测性能较好，能有效降低检测错误率。

英文摘要：

For image steganography,the existing feature extracting method can hardly effectively re- flect the small changes in the images. A local texture feature extraction scheme in multi-layer is pro- posed. First, image is processed by high-pass filtering and segmented based on texture complexity of the image block. Then, for different types of sub-images, the improved local binary pattern （LBP） characteristic is extracted. The final features is cbtained by using fisher discrimination. Experiments are made for LSBM , HUGO and S-UNIWARD steganographic algorithm with embedded at differ- ent rates. Results show that this method take more times, but has good steganalysis detection per- formance and can reduce the detection error rate effectively.