一种基于游程长度和隐写分析特征融合的图像拼接盲检测方法  

An image splicing blind detection method combining run-length with steganalysis feature

作　　者：李佳馨 高铁杠[1] 赖云霓 

机构地区：[1]南开大学软件学院,天津300071

出　　处：《光电子．激光》2015年第7期1387-1393,共7页Journal of Optoelectronics·Laser

基　　金：天津市自然科学基金(11JCZDJC16000)资助项目

摘　　要：提出了一种基于游程长度(RLRN)和隐写分析特征融合的图像拼接检测算法。算法中的隐写分析特征是在图像经分块离散余弦变换(DCT)后的系数矩阵中提取,并将其和RLRN特征进行融合。特征提取在色度(chroma)空间进行,用支持向量机(SVM)作为分类器。实验结果显示,融合后的特征在图像测试库CASIA v1.0和CASIA v2.0上识别率分别达到98.57%和97.27%,不仅比特征在融合前的识别率有较大提高,而且和现有的一些算法相比,提出的特征融合算法也具有良好的识别性能。With the extensive use of editing software,digital image tampering becomes very easy,where the image splicing is the most common.Meanwhile,some images such as evidence in court are very important.It＇s crucial to provide some reliable methods to identify whether an image has been forged.An image splicing detection method combining run-length with steganography analysis feature is proposed.First,the steganalysis feature is extracted by applying a submodel named＂s2_spam12hv＂（apart of the rich model proposed by Fridrich et al）into the coefficients matrix generated by block discrete cosine transform（BDCT）of an image.Then,this feature is combined with the run-length feature.The runlength feature consists of four gray level run-length run-number vectors extracted in four different directions from a de-correlated image.The feature extractions of the two parts are both carried out in chroma space which consists of cb and cr channels.Support vector machine is chosen as the classifier.Experimental results show that the merged feature can achieve accuracies of 98.57%and 97.27%in datasets CASIA v1.0and CASIA v2.0,respectively.The recognition rate of the feature without merging is greatly improved,and the proposed feature fusion algorithm also shows good recognition performance compared with some existing algorithms.

关 键 词：图像拼接检测 分块离散余弦变换(BDCT) 游程长度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]