CAJ | 학술논문

为进一步提高隐写分析中校准特征对嵌入的敏感性，通过分析校准技术及其与特征之间的关系，在校准已有的分类基础上建立了一种数学模型，提出了一种基于该校准特性的JPEG通用隐写分析算法。算法采用剪切4像素的校准操作，结合微分知识提出校准的改进形式，根据校准前后图像特征的空间分布，得到直方图特征；再由冗余关系计算新校准表示下的马尔可夫转移概率矩阵，最后与块间特征融合后得到新特征集。通过对nsF5、Jsteg和MB1算法在较低嵌入率时的检测和新特征集子集比较实验，发现该方法较现有校准分析方法具有更好的检测性能，达到了90％以上的正确率；各特征集也表现了一定的互补性；在不同质量因子实验中性能较为稳定，可靠性较好。
위진일보제고은사분석중교준특정대감입적민감성，통과분석교준기술급기여특정지간적관계，재교준이유적분류기출상건립료일충수학모형，제출료일충기우해교준특성적JPEG통용은사분석산법。산법채용전절4상소적교준조작，결합미분지식제출교준적개진형식，근거교준전후도상특정적공간분포，득도직방도특정；재유용여관계계산신교준표시하적마이가부전이개솔구진，최후여괴간특정융합후득도신특정집。통과대nsF5、Jsteg화MB1산법재교저감입솔시적검측화신특정집자집비교실험，발현해방법교현유교준분석방법구유경호적검측성능，체도료90％이상적정학솔；각특정집야표현료일정적호보성；재불동질량인자실험중성능교위은정，가고성교호。
To improve the embedding sensibility of calibrated feature in steganalysis , by studying the relationship between calibration technique and feature , a mathematical model for calibration based on the calibration classification is established , and a blind JPEG steganalysis algorithm based on the new calibration is presented . First we crop 4 pixels in the image and put forward a modified form of calibration , then the histogram characteristic is obtained according to the spatial distribution of the image features before and after calibration , and the Markov transfer probability matrix of the new calibration is calculated on the basis of redundancy . Finally , we fuse these features with the blocks feature and obtain the feature vector . Through the detection experiment of nsF5 , Jsteg and MB1 algorithms with low embedding rates and among the feature vector , it is shown that this method has a better detection performance compared with those existing calibration methods . Its correctrate is more than 90% . The feature sets also show some complementary characteristics . It can be more stable and reliable in the different quality factors experiment .