Detection of doctored images using bispectral analysis

In this paper we solve the problem of image doctoring using bispectral analysis and Expectation-Maximization (EM) algorithms. Doctoring is a process of modifying or changing the content of an image. Now-a-days there is widespread availability of multi-media data in digital form. The technology that could alter the digital data is growing at break-neck speed, which imposes challenge to image forensic by hiding the sensitivity of the image, leaving no visual artifacts in order to deceive people, for amusement, to mislead people, an attempt to rewrite history, to exaggerate the situations of war, to customize groundbreaking advances in research, etc... The detection of doctoring finds application in law enforcement, medicine, security, military, etc... For the detection of image doctoring, we use bispectral analysis for capturing the optical low-pass property of cameras. We compute the bicoherence magnitude and phase responses for the image. Analysis shows that doctoring operation introduces higher-order frequencies in correlation with lower-order frequencies, which increases magnitude response and phase bias towards ±90° and we use the variation in magnitude and phase bias in the detection process. On an average, the detection of doctoring using bispectral analysis gives a detection rate of 60% for spliced set of images. We also propose Expectation-Maximization (EM) based algorithm to detect the process of doctoring in cloned and re-touched images. We demonstrate the proposed algorithm on different doctored images, which include doctoring using splicing, cloning and re-touching.