An Improved Algorithm For Digital Image Authentication and Forgery Localization Using Demosaicing Artifacts - Université de technologie de Troyes Accéder directement au contenu
Article Dans Une Revue IEEE Access Année : 2019

An Improved Algorithm For Digital Image Authentication and Forgery Localization Using Demosaicing Artifacts

Résumé

This paper focuses on the digital image authentication and forgery localization using demosaicing artifacts. The aim is to build an algorithm allowing a bridge between the color filter array pattern and demosaicing algorithm estimation, and the statistical analysis of demosaicing artifacts in spatial domain to improve the authentication and localization performance. After analyzing the evolution of demosaicing traces in camera acquisition pipeline, a robust feature statistic characterizing demosaiced digital images is first developed on the basis of the noise residue of green channel. Such a feature statistic is less sensitive to the edges problem because only the smooth region of green channel is used in the development. Next, a single normal mixture model is proposed to describe the probability distribution of feature statistics for both original and tampered images. Therefore, normality tests can be used to authenticate automatically digital images. The authentication performance can be further improved by human interpretation of supported graphic tools. Finally, a penalized expectation-maximization algorithm is used to localize forged areas in tampered images. Numerous comparative studies on four well-known datasets show that the developed algorithm yields better performance and robustness than existing forensics algorithms of the same kind.

Dates et versions

hal-02279217 , version 1 (05-09-2019)

Identifiants

Citer

Nhan Le, Florent Retraint. An Improved Algorithm For Digital Image Authentication and Forgery Localization Using Demosaicing Artifacts. IEEE Access, 2019, 7, pp.125038 - 125053. ⟨10.1109/ACCESS.2019.2938467⟩. ⟨hal-02279217⟩
33 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More