基于局部均值分解和矩特征的图像拼接检测

数字图像的真实性检测在信息安全领域有着重要的应用.图像拼接是图像篡改中最常用的手段之一,拼接检测的关键在于提取区分自然图像和篡改图像的有效特征,为此提出了一种基于局部均值分解(LMD)和矩特征的数字图像拼接篡改盲检测算法.该方法将图像检测问题转化为两分类模式识别问题,分别提取图像的LMD特征和矩特征,得到混合特征,并用Adaboost算法进行分类检测.在哥伦比亚大学数字图像拼接检测数据库上的实验结果表明,该算法的检测准确率达到93.61%,优于其他几种常用的检测算法.

Image splicing detection based on local mean decomposition and moment features

The authenticity detection of digital images has important applications in the field of information security.Image splicing is one of the common methods in image tampering.The key of image splicing detection is to extract the effective features of the natural image and the tampered image.This paper presented a digital image splicing blind detection algorithm based on local mean decomposition (LMD) and moment features.This method transforms the image detection problem into two-class pattern recognition problem.It extracted features using LMD and combine the moment features to get mixed features, then used Adaboost as classifier.Experiments were made on Columbia image splicing detection evaluation dataset.It is encouraging to find that the detection accuracy of the algorithm is 93.61%.