A machine learning based technique for detecting digital image resampling

Abstract

Digital images can easily be tampered because of the popularity and power editing software. In order to create a persuasive forged image, the image is usually exposed to several geometric transformations, such as rescaling and rotating. Since the manipulations require a resampling step, uncovering traces of resampling became an important approach for detecting image forgeries. In this paper, we propose a new technique to reveal image resampling artifacts. The technique employs specific features of the linear dependencies of neighboring image samples for discriminating resampled images from original images. A machine learning method is utilized for classification. Experimental results in a large dataset show that the proposed technique is good in detecting resampled images, even when the manipulated images were slightly transformed.