一种改进的SIFT篡改检测算法

摘要

In the process of SIFT ( Scale Invariant Feature Transform ) image registration algorithm, the principal orientation is affected by the dispersion of histogram. Besides,the feature descriptor section of conventional SIFT does not make full use of local feature information. As to these problems,an improved SIFT algorithm on characteristic statistical distributions and consistency constraint will be presented. Firstly,DOG( Difference of Gaussian) scale space feature point detection method is adopted to extract key points. Then, in the process of principal orientation generation,our method selects line with maximum dispersion. Furthermore,this method generates feature descriptor based on characteristic statistical distributions in polar coordinate. Finally, a new matching method based on consistency constraint will be introduced. In experiments,the values of TPR(True Positive Rate) and FPR(False Positive Rate ) is 98. 03% and 7. 99%, we test the performances of our propose method. The experimental results demonstrate the feasibility and effectiveness of our approach.%针对数字图像的复制粘贴盲检测进行了研究,传统的筛选特征描述部分不充分检测图像信息,基于统计分布特征和一致性约束理论提出了一种改进的SIFT篡改检测算法.首先,建立高斯差分(DOC)尺度空间特征点检测方法以提取关键点.然后,在主要方向生成的过程中基于最大色散方法选择,此外,该方法基于统计特征生成特征描述的精确坐标、尺度值、像素区域尺寸值、边界标记、边界角和曲率.最后,基于一致性约束新的匹配方法将介绍.实验结果为真正类率(TPR)值为98.03%,假正类率(FPR)值为7.99%,验证了本文提出的方法的可行性和有效性.该算法对篡改区域的平移、缩放和旋转有较强的鲁棒性.