Exploiting Reference Images in Exposing Geometrical Distortions

Abstract

Nowadays, image alteration in the mainstream media has become common. The degree of manipulation is facilitated by image editing software. In the past two decades the number indicating manipulation of images rapidly grows. Hence, there are many outstanding images which have no provenance information or certainty of authenticity. Therefore, constructing a scientific and automatic way for evaluating image authenticity is an important task, which is the aim of this paper. In spite of having outstanding performance, all the image forensics schemes developed so far have not provided verifiable information about source of tampering. This paper aims to propose a different kind of scheme, by exploiting a group of similar images, to verify the source of tampering. First, we define our definition with regard to tampered image. The distinctive features are obtained by exploiting Scale- Invariant Feature Transform (SIFT) technique. We then proposed clustering technique to identify the tampered region based on distinctive keypoints. In contrast to k-means algorithm, our technique does not require the initialization of k value. The experimental results over and beyond the dataset indicate the efficacy of our proposed scheme.