An improved parameter estimation scheme for image modification detection based on DCT coefficient analysis

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, Q(1), and the portion of the modified region, alpha, have to be estimated, and more accurate estimations of alpha and Q(1) lead to better detection and localization results. Existing methods estimate alpha and Q(1) in a completely blind manner, without considering the characteristics of the mixture model and the constraints to which a should conform. In this paper, we propose a more effective scheme for estimating alpha and Q(1), based on the observations that, the curves on the surface of the likelihood function corresponding to the mixture model is largely smooth, and alpha 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. (C) 2015 Elsevier Ireland Ltd. All rights reserved.