Smoothing Filtering Detection for Digital Image Forensics

In the past few years, as a type of image authentication technique without relying on pre-registration or pre-embedded information, the passive blind image forensics has become a hot issue in the field of information security techniques. In this paper, a novel algorithm for detecting smoothing filtering in digital images is proposed based on the frequency residual. The suspected image is re-filtered with a Gaussian low-pass filter, and the difference between the initial image and the re-filtered image in Fourier domain is called the frequency residual. Then, the frequency residual is projected into the Radon space with an adaptation of Radon transform. The obtained data is modeled as Fourier series and the model parameters are adopted as features for filtering detection. The experimental results show that the proposed algorithm can not only detect three typical smoothing spatial filters, including Gaussian filter, average filter, and median filter, but also can predict parameters of these filters to complement the existing state-of-the-art methods.