摘要
JPEG图像的双量化效应是检测和发现JPEG图像篡改的重要线索。针对现有检测算法大多数是基于DCT块效应而较少利用双量化效应的情况,提出了一种利用JPEG双量化效应的图像篡改盲检测新方案。该方案对比用DCT系数直方图计算的未篡改区域后验概率和用区间长度计算的未篡改区域后验概率之间的差异性,提取能有效区分篡改块和未篡改块的特征,计算每个DCT块的特征值,然后设置阈值,将特征值大于阈值的图像块判定为篡改块,最后通过选取连通区域,标定篡改区域。实验结果表明,与已有的类似方案相比,该方案能够较精确地检测和定位篡改区域,且对颜色、纹理丰富的图像具有明显优势。
The double quantization effect in JPEG images is becoming important clue in detecting image forgeries. As many detection methods are based on the block artifacts of DCT coefficients, a new blind detection scheme which makes use of the double quantization effect is proposed. By comparing two posteriori probabilities of untampered region, which are computed by histogram of DCT coefficients and interval length respectively, a feature that can distinguish the tampered block and untampered block is extracted, and then the feature value of each DCT block is calculated. After setting a threshold, the tampered blocks whose feature values are larger than the threshold are obtained. Finally, the tampered region is located by labeling the connected components. Experimental results demonstrate that the proposed scheme can detect and localize the tampered areas more accurately with respect to previously proposed methods, especially for colorful or rich texture images.
出处
《计算机工程与应用》
CSCD
2014年第20期202-206,共5页
Computer Engineering and Applications
基金
福建省计算机应用技术和信号与信息系统研究生教育创新基地资助(闽高教[2008]114号)
关键词
双量化效应
盲检测
直方图
后验概率
连通区域
double quantization effect
blind detection
histogram
posteriori probability
connected region
