Demystifying histograms of multi-quantised DCT coefficients
Demystifying histograms of multi-quantised DCT coefficients
复制标题
DOI：
10.1109/icme.2011.6012205
复制
发表时间：
2011-07
期刊：
影响因子：
--
通讯作者：
Thomas Gloe
中科院分区：
文献类型：
--
作者：
Thomas Gloe
JPEG 压缩伪影是数字图像取证分析中最突出的特征之一。在这里，我们重点关注单量化和多量化 DCT 系数的直方图的频谱分析。我们提出了有限支持直方图的频率响应模型，它以合理的精度解释了特征量化峰值的位置和大小，并为一般量化伪影的形成提供了有价值的见解。我们推导出一组用于有效分析单压缩和多重压缩图像的实用规则，并讨论在量化表估计中的应用。
JPEG compression artefacts are among the most prominent features in the forensic analysis of digital images. Here, we focus on the spectral analysis of histograms of single- and multi-quantised DCT coefficients. We present a model of the frequency response of finite-support histograms, which explains position and magnitude of characteristic quantisation peaks with reasonable precision and gives valuable insights into the formation of quantisation artefacts in general. We derive a set of practical rules for the efficient analysis of single- and multi-compressed images and discuss applications in the estimation of quantisation tables.