Nowadays, Deepfake videos are widely spread over the Internet, which severely impairs the public trustworthiness and social security. Although more and more reliable detectors have recently sprung up for resisting against that new-emerging tampering technique, some challengeable issues still need to be addressed, such that most of Deepfake video detectors under the framework of the supervised mechanism require a large scale of samples with accurate labels for training. When the amount of the training samples with the true labels are not enough or the training data are maliciously poisoned by adversaries, the supervised classifier is probably not reliable for detection. To tackle that tough issue, it is proposed to design a fully unsupervised Deepfake detector. In particular, in the whole procedure of training or testing, we have no idea of any information about the true labels of samples. First, we novelly design a pseudo-label generator for labeling the training samples, where the traditional hand-crafted features are used to characterize both types of samples. Second, the training samples with the pseudo-labels are fed into the proposed enhanced contrastive learner, in which the discriminative features are further extracted and continually refined by iteration on the guidance of the contrastive loss. Last, relying on the inter-frame correlation, we complete the final binary classification between real and fake videos. A large scale of experimental results empirically verify the effectiveness of our proposed unsupervised Deepfake detector on the benchmark datasets including FF++, Celeb-DF, DFD, DFDC, and UADFV. Furthermore, our proposed well-performed detector is superior to the current unsupervised method, and comparable to the baseline supervised methods. More importantly, when facing the problem of the labeled data poisoned by malicious adversaries or insufficient data for training, our proposed unsupervised Deepfake detector performs its powerful superiority.

The illegal use of facial forgery models, such as Generative Adversarial Networks (GAN) synthesized contents, has been on the rise, thereby posing great threats to personal reputation and national security. To mitigate these threats, recent studies have proposed the use of adversarial watermarks as countermeasures against GAN, effectively disrupting their outputs. However, the majority of these adversarial watermarks exhibit very limited defense ranges, providing defense against only a single GAN forgery model. Although some universal adversarial watermarks have demonstrated impressive results, they lack the defense scalability as a new-emerging forgery model appears. To address the tough issue, we propose a scalable approach even when the original forgery models are unknown. Specifically, a watermark expansion scheme, which mainly involves inheriting, defense and constraint steps, is introduced. On the one hand, the proposed method can effectively inherit the defense range of the prior well-trained adversarial watermark; on the other hand, it can defend against a new forgery model. Extensive experimental results validate the efficacy of the proposed method, exhibiting superior performance and reduced computational time compared to the state-of-the-arts.

The face manipulation technique such as Deepfake has been widely used to create realistic faces, which raises growing concerns in the community. Based on the correct labeled data, the current Deepfake detectors are mostly trained on the clean dataset, usually resulting in the reliable high detection accuracy. However, in the real-world scenario, labelers possibly mislabel the data or malicious attackers always intend to poison the training data with incorrect label, namely noisy label attack, leading to poor detection results. To overcome the tough issue, we propose a Deepfake detection framework fighting against noisy label attack. Specifically, a Negative Sample Generator (NSG) utilizes the possibly-poisoned samples to generate labelreliable negative samples through simulating blending artifacts caused by Deepfake. Next, a Noise-immune Contrastive Learner (NiCL) takes both positive and negative samples as training data, exploring blending artifacts and intrinsic forgery clues to filtrate the noisy samples out. Moreover, relying on label purification, the filtrated noisy samples are further purified, which then are fed back to the feature extractor for the following model training. Extensive experiments on the benchmark datasets demonstrate the superiority of our proposed Deepfake detector. In particular, when fighting against noisy label attack, the high performance of the proposed detector is remarkably better than its competitors.

This paper presents a new change detection technique for images taken from the sentinel-2 satellite between 2015 and 2018 in different regions of the world. These images are widely used in recent years for change detection. This technique is based on two dissimilarity measures: the Disjoint Information and the Local Dissimilarity Map. The disjoint information quantifies the dissimilarities between textures and the Local Dissimilarity Map those between structures of images. Firstly, the disjoint information is computed across the blocks of the RGB image channels and the value is multiplied by the center value of the pixel of each block. Secondly, the Local Dissimilarity Maps over the pre-processed channels and the average of the pixel values on the Local Dissimilarity Maps are computed. Finally, an extension of the Gaussian OTSU’s threshold is used to detect changes in images. Experimental results on the well-known Onera Satellite Change Detection (OSCD) dataset show the effectiveness of our proposed method compared to the state-of-the-art deep learning methods.

Considering the unavailability of labeled data sets in remote sensing change detection, this letter presents a novel and low complexity unsupervised change detection method based on the combination of similarity and dissimilarity measures: Mutual Information (MI), Disjoint Information (DI) and Local Dissimilarity Map (LDM). MI and DI are calculated on sliding windows with a step of 1 pixel for each pair of channels of both images. The resulting scalar values, weighted byqandmcoefficients, are multiplied by the values of the center pixels of the windows weighted bypto remove the textures on images. The changes are detected using respectively the grayscale LDM and color LDM. A sliding window is then used on the color LDM and each pixel is characterized by a two-parameter Weibull distribution. Binarized change maps can be obtained by using ak-means clustering on the model parameters. Experiments on optical aerial image data set show that the proposed method produces comparable, even better results, to the state-of-the-art methods in terms of Recall, Precision and F-measure.

To address the camera origin identification problem of inquiry images, many forensic methods have been proposed. However, the heavy computational overhead and the potential threat of privacy leakage for inquiry images make many existing forensic methods less applicable. Only a few research works have proposed secure forensic methods to address the aforementioned issues; however, they did not give a detailed analysis for the statistical performance. In this paper, we propose an efficient privacy-preserving forensic method with analytical statistical performance to solve the camera model identification problem efficiently and securely. To preserve the privacy of inquiry images, we propose a hybrid privacy-preserving scheme consisting of two operations:Position Scrambling Encryptionto preserve the privacy of image content andNoise Linear-Mapping Processingto preserve the privacy of camera model identity for inquiry images. In the encrypted domain where the proposed privacy-preserving scheme is employed, we first propose a novel statistical noise model, which can accurately characterize an encrypted JPEG inquiry image. Then, a noise model-based detector is designed to identify different camera models. Experimental results verify the feasibility of our proposed method from both privacy-preserving and forensic effectiveness and report that our method outperforms the state-of-the-art secure forensic methods, especially when sample images used to estimate camera fingerprints are insufficient, such as only 2 available images.

Copy-Move forgery has been widely studied as it is a really common forgery. Furthermore, it is the easiest forgery to create with serious security-related threats in particular for distant remote id onboarding where company ask their customer to send a photo of their ID document. It is then easy for a counterfeit to alter the information on the document by copying and pasting letters within the photo. On the other hand, copy-move detection algorithms are known to perform worse in presence of similar but genuine objects preventing us from using them in practical situations like remote ID on boarding. In this article we propose a novel copy-move public dataset containing forged ID documents and study current state-of-the-art performances on this dataset to evaluate their potential use in practical situations.

International audience

The goal of this paper is to propose an accurate method for estimating quantization steps from an image that has been previously JPEG-compressed and stored in lossless format. The method is based on the combination of the quantization effect and the statistics of discrete cosine transform (DCT) coefficient characterized by the statistical model that has been proposed in our previous works. The analysis of quantization effect is performed within a mathematical framework, which justifies the relation of local maxima of the number of integer quantized forward coefficients with the true quantization step. From the candidate set of the true quantization step given by the previous analysis, the statistical model of DCT coefficients is used to provide the optimal quantization step candidate. The proposed method can also be exploited to estimate the secondary quantization table in a double-JPEG compressed image stored in lossless format and detect the presence of JPEG compression. Numerical experiments on large image databases with different image sizes and quality factors highlight the high accuracy of the proposed method.

This paper investigates the problem of identifying the source imaging device of the same model for a natural raw image. The approach is based on the Poissonian-Gaussian noise model which can accurately describe the distribution of the given image. This model relies on two parameters considered as unique fingerprint to identify source cameras of the same model. The identification is cast in the framework of hypothesis testing theory. In an ideal context where all model parameters are perfectly known, the Likelihood Ratio Test (LRT) is presented and its performance is theoretically established. The statistical performance of LRT serves as an upper bound of the detection power. For a practice use, when the image parameters are unknown and camera parameters are known, a detector based on estimation of those parameters is designed. Numerical results on simulated data and real natural raw images highlight the relevance of our proposed approach.

This paper investigates the statistical detection of JSteg steganography. The approach is based on a statistical model of discrete cosine transformation (DCT) coefficients challenging the usual assumption that among a subband all the coefficients are independent and identically distributed (i. i. d.). The hidden information-detection problem is cast in the framework of hypothesis testing theory. In an ideal context where all model parameters are perfectly known, the likelihood ratio test (LRT) is presented, and its performances are theoretically established. The statistical performance of LRT serves as an upper bound for the detection power. For a practical use where the distribution parameters are unknown, by exploring a DCT channel selection, a detector based on estimation of those parameters is designed. The loss of power of the proposed detector compared with the optimal LRT is small, which shows the relevance of the proposed approach.

This paper investigates the statistical detection of Jsteg steganography. The approach is based on the statistical model of Discrete Cosine Transformation (DCT) coefficients. The hidden information detection problem is cast in the framework of hypothesis testing theory. In an ideal context where all model parameters are perfectly known, the Likelihood Ratio Test (LRT) is presented and its performances are theoretically established. The statistical performance of LRT serves as an upper bound of the detection power. For a practical use, when the distribution parameters are unknown, a detector based on estimation of those parameters is designed. The loss of power of the proposed detector, compared with the optimal LRT is small, which shows the relevance of the proposed approach.

This paper investigates the statistical detection of data hidden within DCT coefficients of JPEG images using a Laplacian distribution model. The main contributions is twofold. First, this paper proposes to model the DCT coefficients using a Laplacian distribution but challenges the usual assumption that among a sub-band all the coefficients follow are independent and identically distributed (i.i.d). In this paper it is assumed that the distribution parameters change from DCT coefficient to DCT coefficient. Second this paper applies this model to design a statistical test, based on hypothesis testing theory, which aims at detecting data hidden within DCT coefficient with the JSteg algorithm. The proposed optimal detector carefully takes into account the distribution parameters as nuisance parameters. Numerical results on simulated data as well as on numerical images database show the relevance of the proposed model and the good performance of the ensuing test.

The goal of this paper is to design a statistical test for the camera model identification problem. The approach is based on the state-of-the-art model of Discret Cosine Transform (DCT) coefficients to capture their statistical difference, which jointly results from different sensor noises and in-camera processing algorithms. The noise model parameters are considered as camera fingerprint to identify camera models. The camera model identification problem is cast in the framework of hypothesis testing theory. In an ideal context where all model parameters are perfectly known, this paper studies the optimal detector given by the Likelihood Ratio Test (LRT) and analytically establishes its statistical performances. In practice, a Generalized LRT is designed to deal with the difficulty of unknown parameters such that it can meet a prescribed false alarm probability while ensuring a high detection performance. Numerical results on simulated database and natural JPEG images highlight the relevance of the proposed approach.

The goal of this paper is to design a statistical test for the camera model identification problem. The approach is based on the heteroscedastic noise model, which more accurately describes a natural raw image. This model is characterized by only two parameters, which are considered as unique fingerprint to identify camera models. The camera model identification problem is cast in the framework of hypothesis testing theory. In an ideal context where all model parameters are perfectly known, the likelihood ratio test (LRT) is presented and its performances are theoretically established. For a practical use, two generalized LRTs are designed to deal with unknown model parameters so that they can meet a prescribed false alarm probability while ensuring a high detection performance. Numerical results on simulated images and real natural raw images highlight the relevance of the proposed approach.

The goal of this paper is to propose a statistical model of quantized discrete cosine transform (DCT) coefficients. It relies on a mathematical framework of studying the image processing pipeline of a typical digital camera instead of fitting empirical data with a variety of popular models proposed in this paper. To highlight the accuracy of the proposed model, this paper exploits it for the detection of hidden information in JPEG images. By formulating the hidden data detection as a hypothesis testing, this paper studies the most powerful likelihood ratio test for the steganalysis of Jsteg algorithm and establishes theoretically its statistical performance. Based on the proposed model of DCT coefficients, a maximum likelihood estimator for embedding rate is also designed. Numerical results on simulated and real images emphasize the accuracy of the proposed model and the performance of the proposed test.

In this paper a new methodology for detecting anomaly from few tomography projections is presented. This methodology exploits a statistical model adapted to the content of radiographs together with hypothesis testing theory. The main contributions are the following. First, using a generic model of the tomography acquisition pipeline, the whole non-destructive testing process is entirely automated. Second, by using testing theory the statistical properties of the proposed test are analytically established. This particularly permits the guaranteeing of a prescribed false-alarm probability and allows us to show that the proposed test is almost optimal. Experimental results show the sharpness of the established results and the relevance of the methodology.

This paper proposes a novel method, based on hypothesis testing theory, to detect data hidden with the LSB matching. When all the image parameters, a test which asymptotically maximizes the detection power and guarantees a false-alarm probability, is presented and its statistical properties are analytically given in a closed-form. This provides an asymptotic upper-bound for the power of any detector for LSB matching. In practice the image parameters are unknown. A Generalized Likelihood Ratio Test (GLRT) is proposed and its statistical properties are also analytically established. Numerical results and comparisons with prior art detectors highlight the relevance of the proposed methodology.

The goal of the paper is to propose an optimal statistical test for the steganalysis of Jsteg algorithm. The test is based on a state-of-the-art statistical model of quantized Discrete Cosine Transform (DCT) coefficients that allows us to reliably detect any small change in a cover image due to hidden information. By formulating the hidden information detection as a hypothesis testing problem, the paper designs the most powerful Likelihood Ratio Test (LRT) assuming that all model parameters are perfectly known. The statistical performance of the LRT is analytically provided. Numerical results and comparison with other detectors highlight the relevance of the proposed approach.

This paper investigates the detection of information hidden in digital media by the least significant bit (LSB) matching scheme. In a theoretical context of known medium parameters, two important results are presented. First, based on the likelihood ratio test, we present a test that asymptotically maximizes the detection power whatever the embedding rate might be. Second, the statistical properties of this test are analytically calculated; it is particularly shown that the decision threshold which warrants a given probability of false-alarm is independent of inspected medium parameters. This provides an asymptotic upper-bound for the detection power of any test that aims at detecting data hidden with the LSB matching method. In practice, when detecting LSB matching, the unknown medium parameters have to be estimated. Based on a local model of digital media, a generalized likelihood ratio test is presented by replacing the unknown parameters by their estimation. Numerical results on large databases highlight the relevance of the proposed methodology and comparison with state-of-the-art detectors shows that the proposed tests perform well.

The goal of this paper is to propose the optimal statistical test based on the modeling of discrete cosine transform (DCT) coefficients with a quantified Laplacian distribution. This paper focuses on the detection of hidden information embedded in bits of the DCT coefficients of a JPEG image. This problem is difficult, in terms of statistical decision, for two main reasons: first, the number of DCT coefficients used to conceal the hidden bits is random; second, the JPEG image compression induces a strong quantization of DCT coefficients. The proposed test explicitly takes into account the randomness of the number of DCT coefficients used. It maximizes the probability of hidden information detection by ensuring a prescribed level of false alarm.

In the last two decades substantial progress has been made in the detection of hidden information or hidden communication channels in media files or streams. Typically, it is necessary to reliably detect in a huge set of files (image, audio, and video) which of these files contain the hidden information. The goal of this paper is to study the problem of hypothesis testing based on quantized observations by using a parametric statistical model with nuisance parameters and to apply the obtained tests to the hidden information detection.

The non-Bayesian detection of an anomaly from a single or a few noisy tomographic projections is considered as a statistical hypotheses testing problem. It is supposed that a radiography is composed of an imaged nonanomalous background medium, considered as a deterministic nuisance parameter, with a possibly hidden anomaly. Because the full voxel-by-voxel reconstruction is impossible, an original tomographic method based on the parametric models of the nonanomalous background medium and radiographic process is proposed to fill up the gap in the missing data. Exploiting this "parametric tomography," a new detection scheme with a limited loss of optimality is proposed as an alternative to the nonlinear generalized likelihood ratio test, which is untractable in the context of nondestructive testing for the objects with uncertainties in their physical/geometrical properties. The theoretical results are illustrated by the processing of real radiographies for the nuclear fuel rod inspection.

The paper concerns the radiographic non-destructive testing of well-manufactured objects. The detection of anomalies is addressed from the statistical point of view as a binary hypothesis testing problem with nonlinear nuisance parameters. A new detection scheme is proposed as an alternative to the classical GLR test. It is shown that this original decision rule detects anomalies with a loss of a negligible (epsiv) part of optimality with respect to an optimal invariant test designed for the "closest" hypothesis testing problem with linear nuisance parameters