【作者】 刘绍辉； 龚龙； 姚鸿勋；

【Author】 Liu Shaohui,Gong Long,and Yao Hongxun (School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001)

【机构】 哈尔滨工业大学计算机科学与技术学院；

【摘要】 隐写分析是信息安全领域一个新的研究热点,其中多数为针对特定隐藏算法的隐写分析算法,少数泛盲隐写分析算法又具有复杂度高正检率低的缺点.对图像像素值扣除受隐藏改变很小的高位后剩余的低位图像进行小波变换,恢复载体图像,利用梯度能量之差形成12维特征向量,最后通过支持向量机(SVM)进行训练分类.在3000幅训练图像库和3000幅测试图像库上(没有交集),分别对LSB(least significant bits)、自适应空域和BPCS(bit-plane complexity segmentation)等多种空域隐藏算法进行训练和测试.实验结果表明,算法有很好的检测性能,载密图像和载体图像的平均正检率分别为93.7%和96.2%.更多还原

【Abstract】 Steganslysis is becoming a hot research issue of information security recently,most of which aims at specific steganography,and some universal blind steganalysis algorithms have the drawback of high complexity and low correct detecting ratio.In this paper,firstly approximated cover images are estimated by using wavelet transform de-noising method,then 12 feature vectors are obtained by calculating the differences of gradient energy between stego and estimated cover images. Finally,SVM is employed to train and test those feature vectors.In experiments,training and testing images databases include 6000 images(3000 for each and not overlapped),of which stego images are generated by several well-known steganography algorithms in spatial domain including LSB,adaptive spatial hiding and BPCS methods.Experimental results verify the performance of the proposed method;average correct detecting ratio for stego and cover images is 93.7%and 96.2 respectively.更多还原