Research On Blind Forensics Technology Of Splicing Image Based On Interpolation

At present, the digital image is facing the threat of tampering and forgery at will, which makes the concepts of "the photo is evidence" and "seeing is believing" completely broken. Therefore, it is very important to explore and understand the authenticity and credibility of the image. The conventional active forensics technology such as digital watermarking and so on needs to pre-process the images, which makes the application of this technology have some limitations in practice. Therefore, this thesis studies a forensic technology, digital image blind forensics, which does not rely on identification information.Image resampling is widely used in the processes of image acquisition and image scaling and so on. Image interpolation plays an important role in these processes. In addition, image splicing is one of the most commonly used techniques in digital image tampering. Therefore, this thesis focuses on the problem of authenticity detection in digital image blind forensics. The theory and technology of blind forensics for splicing image based on interpolation are deeply studied.Firstly, the principle of color image is introduced in details in this thesis. Bayer-patterned color filter array (CFA) and its interpolation principle are mainly studied. The interpolation methods commonly used in image processing are explored and the patterns of signals interpolated by different methods, which provides a theoretical basis for the work of image blind forensics.In order to further verify the characteristic of image interpolation signal, the periodicity is derived, which exists in the variance and covariance of the derivative. Image interpolation is detected based on periodicity and Radon transform. Proposed algorithm can be used to detect the region of interest or the whole image. Simulation experiments show that proposed algorithm can detect the images which come from different sources, the images processed by different interpolation algorithms and different image processing software. With circling the suspicious area manually, interpolation operation of the tampered image can be detected.To further realize the automatic detection, localization of tampering area and improvement of the detection accuracy rate, the splicing image blind forensics algorithm based on bi-cubic interpolation and Gaussian mixture model (GMM) is proposed. In order to reduce the prediction error, the original image is estimated by bi-cubic interpolation. The characteristics of the mean and variance of the interpolation signal are studied, and the feature based on the variance of prediction error is extracted. GMM is used to model the feature at different locations and the parameters in the model are estimated using expectation maximization (EM) algorithm. Simulation experiments show that proposed algorithm in this thesis can be used to automatically locate the tampering area of the splicing image successfully and improve the detection accuracy rate.