Dual tree complex wavelet transform approach to copy-rotate-move forgery detection

Abstract

As a common scheme of image tampering, copy-move forgery plays an important role in image forgery committee. Although several blind methods aimed at detecting replicated regions have been proposed, these methods cannot detect the rotation changes in the duplicated areas efficiently. In this paper, a new forensic method is presented to detect the replicated areas rotated by arbitrary angles, even by JPEG compression. To achieve this, overlapping blocks of pixels are decomposed using dual tree complex wavelet transform (DTCWT), and then channel energies are extracted from each subband at each decomposition level using the L₁ norm. Finally, the anisotropic rotationally invariant features are extracted using magnitudes of discrete Fourier transform for these channel energies. The implementation procedure of the proposed method is described in detail. Extensive experimental results, including a comparative evaluation with existing methods and the special applications in practice, are also presented to demonstrate the robustness and effectiveness of the proposed method.