A classifier framework for the detection of doctored images

In this paper, we propose a classifier framework for the detection of doctored images. Doctoring is a process of altering or modifying the contents of an authentic image with varied motivations. The advances in technology has made it easy to doctor an authentic image in several ways, and because the images are used as evidences in many areas, identifying the authenticity of an image is of significant importance. We provide a feature set to capture the characteristics of an authentic image and also present a detection methodology to detect a doctored image using a variant of the Bayesian classifier. We model the imaging system and the doctoring process together as a linear system and extract features of doctoring. We show a variant of the Bayesian classifier for the classification of a given image as authentic or doctored and also show that the proposed method gives comparable results with the SVM classifier. We demonstrate the proposed framework for the detection of three different types of doctored images: spliced, cloned and retouched images and achieve an overall accuracy of 78.33% with 9 features, which is higher than the reported accuracy in the literature.