David Vazquez-Padin

Detection of video double encoding with GOP size estimation

Video forensics is an emerging discipline, that aims at inferring information about the processing history undergone by a digital video in a blind fashion. In this work we introduce a new forensic footprint and, based on it, propose a method for detecting whether a video has been encoded twice; if this is the case, we also estimate the size of the Group Of Pictures (GOP) employed during the first encoding. As shown in the experiments, the footprint proves to be very robust even in realistic settings (i.e., when encoding is carried out using typical compression rates), that are rarely addressed by existing techniques.

Prefilter design for forensic resampling estimation

Starting from a theoretical analysis of the resampling estimation problem for image tampering detection, this work presents a study, based on cyclostationarity theory, about the use of prefilters to improve the estimation accuracy of the resampling factor. Considering the methods that perform the estimation by analyzing the spectrum of the covariance of a resampled region, we propose an analytical framework that allows the definition of a cost function that measures the degree of detectability of the spectral peaks. Based on this measure, the design of the optimum prefilters for a particular resampling factor can be solved numerically. Experimental results validate the developed analysis and illustrate the enhancement of the performance in a real scenario.

Two-dimensional statistical test for the presence of almost cyclostationarity on images

In this work, we study the presence of almost cyclostationary fields in images for the detection and estimation of digital forgeries. The almost periodically correlated fields in the two-dimensional space are introduced by the necessary interpolation operation associated with the applied spatial transformation. In this theoretical context, we extend a statistical time-domain test for presence of cyclostationarity to the two-dimensional space. The proposed method allows us to estimate the scaling factor and the rotation angle of resized and rotated images, respectively. Examples of the output of our method are shown and comparative results are presented to evaluate the performance of the two-dimensional extension.

Localization of forgeries in MPEG-2 video through GOP size and DQ analysis

This work addresses forgery localization in MPEG-2 compressed videos. The proposed method is based on the analysis of Double Quantization (DQ) traces in frames that were encoded twice as intra (i.e., I-frames). Employing a state-of-the-art method, such frames are located in the video under analysis by estimating the size of the Group Of Pictures (GOP) that was used in the first compression; then, the DQ analysis is devised for the MPEG-2 encoding scheme and applied to frames that were intra-coded in both the first and second compression. In such a way, regions that were manipulated between the two encodings are detected. Compared to existing methods based on double quantization analysis, the proposed scheme makes forgery localization possible on a wider range of settings.

ML estimation of the resampling factor

In this work, the problem of resampling factor estimation for tampering detection is addressed following the maximum likelihood criterion. By relying on the rounding operation applied after resampling, an approximation of the likelihood function of the quantized resampled signal is obtained. From the underlying statistical model, the maximum likelihood estimate is derived for one-dimensional signals and a piecewise linear interpolation. The performance of the obtained estimator is evaluated, showing that it outperforms state-of-the-art methods.

An SVD approach to forensic image resampling detection

This paper describes a new strategy for image resampling detection whenever the applied resampling factor is larger than one. Delving into the linear dependencies induced in an image after the application of an upsampling operation, we show that interpolated images belong to a subspace defined by the interpolation kernel. Within this framework, by computing the SVD of a given image block and a measure of its degree of saturated pixels per row/column, we derive a simple detector capable of discriminating between upsampled images and genuine images. Furthermore, the proposed detector shows remarkable results with blocks of small size and outperforms state-of-the-art methods.

Set-membership identification of resampled signals

The problem of resampling factor estimation as a means for tampering detection has been largely investigated. Most of the existing techniques rely on the analysis of cyclic correlations induced in the resampled signal. However, in this paper, a new direction is explored by addressing the same problem in terms of the set-membership estimation theory. The proposed technique constructs a model of the problem using available a priori knowledge and in consonance with a finite number of observations that comes from the resampled signal under study. With this information, the proposed technique is able to provide an estimate of the resampling factor applied to the original signal and, if required, an estimate of such signal and an estimate of the interpolation filter. The performance in terms of accuracy and MSE of the proposed approach is evaluated and comparative results with state-of-the-art methods are reported.

Exposing original and duplicated regions using SIFT features and resampling traces

A common type of digital image forgery is the duplication of a region in the same image to conceal something in a captured scene. The detection of region duplication forgeries has been recently addressed using methods based on SIFT features that provide points of the regions involved in the tampering and also the parameters of the geometric transformation between both regions. However, considering this output, there is not yet any information about which of the regions are originals and which are the duplicated ones. A reliable image forensic analysis must provide this information. In this paper, we propose to use a resampling-based method to provide an accurate way to distinguish the original and the tampered regions by analizing the resampling factor of each area. Comparative results are presented to evaluate the performance of the combination of both methods.

A Random Matrix Approach to the Forensic Analysis of Upscaled Images

The forensic analysis of resampling traces in upscaled images is addressed via subspace decomposition and random matrix theory principles. In this context, we derive the asymptotic eigenvalue distribution of sample autocorrelation matrices corresponding to genuine and upscaled images. To achieve this, we model genuine images as an autoregressive random field and we characterize upscaled images as a noisy version of a lower dimensional signal. Following the intuition behind Marčenko-Pastur law, we show that for upscaled images, the gap between the eigenvalues corresponding to the low-dimensional signal and the ones from the background noise can be enhanced by extracting a small number of consecutive columns/rows from the matrix of observations. In addition, using bounds provided by the same law for the eigenvalues of the noise space, we propose a detector for exposing traces of resampling. Finally, since an interval of plausible resampling factors can be inferred from the position of the gap, we empirically demonstrate that by using the resulting range as the search space of existing estimators (based on different principles), a better estimation accuracy can be attained with respect to the standalone versions of the latter.

Random Matrix Theory for Modeling the Eigenvalue Distribution of Images Under Upscaling

The stochastic representation of digital images through a two-dimensional autoregressive (2D-AR) model offers a proper way to approximate the empirical distribution of the eigenvalues coming from genuine images. By considering this model, we apply random matrix theory to analytically derive the asymptotic eigenvalue distribution of causal 2D-AR random fields that have undergone an upscaling operation with a particular interpolation kernel. This eigenvalue characterization is useful in developing new forensic techniques for image resampling detection since we can use theoretical bounds to drive the decision of detectors based on subspace decomposition. Moreover, experimental results with real images show that the obtained asymptotic limits turn out to be excellent approximations, even when working with images of small size.