Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6551881 | Forensic Science International | 2016 | 21 Pages |
Abstract
Most of the existing image modification detection methods which are based on DCT coefficient analysis model the distribution of DCT coefficients as a mixture of a modified and an unchanged component. To separate the two components, two parameters, which are the primary quantization step, Q1, and the portion of the modified region, α, have to be estimated, and more accurate estimations of α and Q1 lead to better detection and localization results. Existing methods estimate α and Q1 in a completely blind manner, without considering the characteristics of the mixture model and the constraints to which α should conform. In this paper, we propose a more effective scheme for estimating α and Q1, based on the observations that, the curves on the surface of the likelihood function corresponding to the mixture model is largely smooth, and α can take values only in a discrete set. We conduct extensive experiments to evaluate the proposed method, and the experimental results confirm the efficacy of our method.
Related Topics
Physical Sciences and Engineering
Chemistry
Analytical Chemistry
Authors
Liyang Yu, Qi Han, Xiamu Niu, S.M. Yiu, Junbin Fang, Ye Zhang,