基于后验概率的图像模糊检测方法  

Blurring detection in image forensics based on posterior probability

作　　者：潘生军[1] 杨本娟[1] 刘本永[1,2] 

机构地区：[1]贵州大学计算机科学与信息学院,贵阳550025 [2]贵州大学智能信息处理研究所,贵阳550025

出　　处：《计算机工程与应用》2012年第32期181-186,共6页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60862003);科技部国际合作项目(No.2009DFR10530);贵州省工业科技攻关项目(No.黔科合GY字(2010)3054);教育部高等院校博士点基金(No.20095201110002);贵州大学研究生创新基金(No.2012035)

摘　　要：数字图像被动盲取证是指在不依赖任何预签名提取或预嵌入信息的前提下,对图像的真伪和来源进行鉴别和取证。图像在经篡改操作时,为了消除图像在拼接边缘产生的畸变,篡改者通常会采用后处理消除伪造痕迹,其中,模糊操作是最常用的手法之一。提出一种人工模糊痕迹检测方法。将经过模糊操作后图像像素之间存在的高度相关性进行模型化表示;采用EM算法估算出图像中每个像素属于上述模型的后验概率;根据所得后验概率的大小进行模糊操作检测。实验结果表明,该算法能够有效地检测出篡改图像中的人工模糊痕迹,并对不同模糊类型、有损JPEG压缩以及全局缩放操作均具有较好的鲁棒性。Digital image passive blind forensic techniques aim to examine the authenticity and sources of digital images without relying on any pre-extraction or pre-embedded information. While the image is tampered, in order to eliminate the visual edge distortion caused by splicing, some post-processing operations are usually employed to eliminate the tampering traces. Among them, blurring operation is one of the most commonly used approaches. In this paper, a novel method which can detect manual blurring in the tampered image is proposed. A model of the high correlation of pixels in an artificially blurred image is proposed. EM algorithm is adopted to estimate the posterior probability that a pixel belongs to this model. The value of the posteriori probability is used for detecting the trace of blurring opera- tion. Experimental results show that this method can effectively detect manual blurring in a tampered image and al- so has a good robustness against different blurring type, lossy JPEG compression, and global scale operation as well.

关 键 词：数字图像盲取证 人工模糊检测 后验概率估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]