Multi-scale local texture descriptor for image forgery detection

In this paper, a multi-scale local texture descriptor is proposed for an image forgery detection framework. In the framework, first, an input image is decomposed into chromatic channel. Then, undecimated wavelet transform is applied to the channel to extract lower subband. Inspired by the Weber's Law, the proposed multi-scale local texture descriptor, called Weber pattern (WP), is calculated from the subband. The WP histogram is considered as the feature of the image. Support vector machine is used as a classifier in the framework. Experimental results on different image datasets show the superiority, in terms of accuracy, of the proposed method over two other contemporary methods in image forgery detection.