Image forgery detection using Markov features in undecimated wavelet transform

Image forgery has become common due to the availability of high-quality image editing softwares. For detecting image forgery there is a need to have important features of the image. For obtaining the image features we need a suitable transform. One of the important and commonly used transform is discrete wavelet transform that can provide spatial and frequency related information of a signal. However, it provides ambiguous information due to its shift variant property. This ambiguity can be overcome using the UWT due to its shift invariance property. In forgery, some operations are applied to an image at different locations. For instance same type of details at different locations, the UWT provides the output of same nature, whereas the DWT doesn't. Due to this property, the features extracted in undecimated wavelet transform (UWT) domain provide better results in many applications like denoising, change detection, etc. In this paper, image features are extracted using the Markov model after transforming it into UWT domain. To evaluate the performance CASIA v1.0, Columbia Color and DSO-1 databases are used. The support vector machine with the linear kernel applied to separate the forged and pristine images. We experimentally obtain better results using the UWT transform as compared to the DWT transform on all these three databases.