Curvelet transform and local texture based image forgery detection

Abstract

In this paper, image forgery detection method based on the curvelet transform and local binary pattern (LBP) is proposed. First, a color image is converted into the chrominance space. Then, the curvelet transform is applied to the chrominance component to decompose it into several scale and orientation wedges. The LBP normalized histogram is calculated from each of the wedges. The final feature vector is obtained by fusing all the histograms. The proposed method is evaluated on three image forgery datasets and compared with some state of the art methods. Experimental results demonstrate the superiority of the proposed method over the compared methods. The detection accuracy of the proposed method is 93.4% 97.0 % and 94.2% on the CASIA TIDE v1.0, CASIA TIDE v2.0 and Columbia color databases, respectively. © 2013 Springer-Verlag.