Abstract
In this paper, a detailed evaluation of multi-scale Weber local descriptors (WLD) based image forgery detection method is presented. Multi-scale WLD extracts the features from chrominance components of an image, which usually encode the tampering information that escapes the human eyes. The WLD incorporates differential excitation and gradient orientation of a center pixel around a neighborhood. In the multi-scale WLD, three different neighborhoods are chosen. A support vector machine is used for classification purpose. The experiments are conducted on three image databases, namely, CASIA v1.0, CASIA v2.0, and Columbia color. The experimental results show that the accuracy rate of the proposed method are 94.19% for CASIA v1.0, 96.61% for CASIA v2.0, and 94.17% for Columbia dataset. These accuracies are significantly higher than those obtained by some state-of-the-art methods.