摘要
计算机生成图像与自然图像在物理生成机理上的差异性导致其在噪声分布上具有明显的区别,据此提出一种基于图像噪声分析的计算机生成图像检测新方法。首先,用小波隐马尔可夫(MMT)模型对图像噪声进行预处理;接着,基于最大似然估计方法估测图像参考模式噪声;最后,在提取图像模式噪声统计距及其与参考模式噪声相关差值基础上,利用支持向量机(SVM)分类器进行鉴别。实验结果显示,较已有的典型算法,该算法具有更好的检测率。
The discrimination of computer generated images from real images becomes more and more important.A novel digital forensics technique to distinguish computer generated images from real images is proposed based on the differences in the noise distribution of images.More specifically,at first,image noise is preprocessed using wavelet-domain hidden Markov tree models,and then the pattern noise is estimated relying on maximum likelihood estimate,finally,feature vectors are extracted such as statistics and differential correlation coefficient which will be classified by SVM.The experiment results show that this method offers a significant improvement in the performance compared with existing typical methods.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2010年第2期293-297,共5页
Journal of Optoelectronics·Laser
基金
国家"863"计划资助项目(2007AA01Z455)
国家自然科学基金资助项目(60772098
60772042)
教育部新世纪优秀人才支持计划资助项目(NCET-0600393)
2007年上海市曙光计划资助项目
