Search

The reliability of multimedia is being progressively tested by sophisticated Image Manipulation localization (IML) methods, which has led to the creation of the IML domain. A good manipulation model requires extracting non-semantic differences features between manipulated and authentic regions to exploit artifacts, which calls for explicit comparisons between the two areas. Existing models either use handcrafted-based feature methods, convolutional neural networks (CNNs), or a combination of both. Handcrafted feature methods assume the tampering beforehand, limiting their capabilities for diverse tampering operations, while CNNs model semantic information, which is not enough for the manipulation artifact. To improve these limitations, we have designed a dual-branch model that combines handcrafted feature noise and CNNs as an Encoder-decoder(ED) powered by the attention mechanism. This dual-branch model uses noise features on one branch and RGB on the other before feeding to an ED architecture for semantic learning and skip connection deployed to retain spatial information. Furthermore, this architecture uses channel spatial attention to strengthen further and refine the features' representation. Extensive experimentation on the shallowfakes dataset (CASIA, COVERAGE, COLUMBIA, NIST16) and deepfake datasets Faceforensics + + (FF + +) to demonstrate the superior feature extraction capabilities and performance to various baseline models with AUC score even reaching 99%. Also, it is one of the first methods to perform localization on the deepfake dataset. The model is relatively lighter, has 38 million parameters, and easily outperforms other State-of-the-Art(SoTA) models. [ABSTRACT FROM AUTHOR]

The emergence of deep learning has sparked notable strides in the quality of synthetic media. Yet, as photorealism reaches new heights, the line between generated and authentic images blurs, raising concerns about the dissemination of counterfeit or manipulated content online. Consequently, there is a pressing need to develop automated tools capable of effectively distinguishing synthetic images, especially those portraying faces, which is one of the most commonly encountered issues. In this work, we propose a novel approach to synthetic face discrimination, leveraging deep learning-based image compression and predominantly utilizing the quality metrics of an image to determine its authenticity. [ABSTRACT FROM AUTHOR]

Source camera identification is a crucial problem in digital forensics, as it can help to determine the origin and credibility of digital images. However, most of the existing source camera identification models are not scalable, meaning that they cannot maintain accuracy level when the number of camera models increases. Moreover, these models are vulnerable to post-processing operations, which are often applied by users or online platforms to modify or enhance images. In this paper, we propose a novel and robust source camera identification model based on convolutional neural networks and photo-response non-uniformity fingerprints. Our model extracts three-channel photo-response non-uniformity fingerprints from 320 × 320 image patches and feeds them to a 7-layer convolutional neural network for camera model classification. Our model achieves state-of-the-art accuracy of 98.36% and 97.33% at the brand and model levels, respectively, on a large and diverse dataset of 27 camera models. Our model does not require any preprocessing steps such as patch selection, multi-classifier, or majority voting and is robust to various image processing attacks such as double JPEG compression, smoothing, sharpening, blurring, and re-scaling. [ABSTRACT FROM AUTHOR]

In an era marked by the contrast between information and disinformation, the ability to differentiate between authentic and manipulated images holds immense importance for both security professionals and the scientific community. Copy-move forgery is widely practiced thus, sprang up as a prevalent form of image manipulation among different types of forgeries. In this counterfeiting process, a region of an image is copied and pasted into different parts of the same image to hide or replicate the same objects. As copy-move forgery is hard to detect and localize, a swift and efficacious detection scheme based on keypoint detection is introduced. Especially the localization of forged areas becomes more difficult when the forged image is subjected to different post-processing attacks and geometrical attacks. In this paper, a robust, translation-invariant, and efficient copy-move forgery detection technique has been introduced. To achieve this goal, we developed an AKAZE-driven keypoint-based forgery detection technique. AKAZE is applied to the LL sub-band of the SWT-transformed image to extract translation invariant features, rather than extracting them directly from the original image. We then use the DBSCAN clustering algorithm and a uniform quantizer on each cluster to form group pairs based on their feature descriptor values. To mitigate false positives, keypoint pairs are separated by a distance greater than a predefined shift vector distance. This process forms a collection of keypoints within each cluster by leveraging their similarities in feature descriptors. Our clustering-based similarity-matching algorithm effectively locates the forged region. To assess the proposed scheme we deploy it on different datasets with post-processing attacks ranging from blurring, color reduction, contrast adjustment, brightness change, and noise addition. Even our method successfully withstands geometrical manipulations like rotation, skewing, and different affine transform attacks. Visual outcomes, numerical results, and comparative analysis show that the proposed model accurately detects the forged area with fewer false positives and is more computationally efficient than other methods. [ABSTRACT FROM AUTHOR]

The existing seam carving tamper detection algorithms have the problems of low detection accuracy and weak robustness for the case of low scaling factor. A seam carving tamper detection algorithm integrated with hybrid attention mechanism is proposed. Firstly, BayarConv2D constrained convolution is used to preprocess the image, fully learn the noise characteristics of the image, and fuse the features with RGB image through matrix multiplication. Then, ResNet is used as the backbone network for feature learning, and the residual propagation and residual feedback mechanisms are introduced to highlight the operation traces of seam carving. Finally, the hybrid attention mechanism is used to simultaneously extract the features between adjacent locations and channels to better capture the global features, and then input them into the full connection layer to achieve classification. The experimental results show that on the BOSSBase1.01 dataset, when the scaling factor is 1% and 9%, the detection accuracy of the proposed method reaches 89.48% and 97.94% respectively, which is better than existing mainstream methods. At the same time, it has lower computational complexity and better robustness, and can resist JPEG compression attacks.

Extensive research has been conducted in image forensics on the analysis of double-compressed images, particularly in the widely adopted JPEG format. However, there is a lack of methods to detect double compression in the HEIF format, which has recently gained popularity since it allows for reduced file size while maintaining image quality. Traditional JPEG-based techniques do not apply to HEIF due to its distinct encoding algorithms. We previously proposed a method to detect double compression in HEIF images based on Farid's work on coding ghosts in JPEG images. However, this method was limited to scenarios where the quality parameter used for the first encoding was larger than for the second encoding. In this study, we propose a lightweight image classifier to extend the existing model, enabling the identification of double-compressed images without heavily depending on the input image's quantization history. This extended model outperforms the previous approach and, despite its lightness, demonstrates excellent detection accuracy. [ABSTRACT FROM AUTHOR]

Detecting sources in digital images and videos is crucial to multimedia forensics research. The inherent physical properties of imaging sensors result in the presence of Photo Response Non-Uniformity Noise (PRNU) within the captured multimedia content. This particular noise, often called the "fingerprint," is a unique and stable feature for identifying the source camera. However, the compression of images by platform codecs on social network media introduces varying degrees of quantization noise, making to address this issue, a noise extractor based on variance stabilizing transform and adaptive block clustering principal component analysis (PCA) is proposed, along with an enhanced processing model that incorporates cyclic residual recycling. Firstly, the GAT and adaptive block clustering PCA filtering are applied to extract noises from the images. The obtained noises are then subjected to zero-mean and diagonal artifact elimination processing. Next, the real and imaginary parts of the noise spectrum are individually subjected to real-time iterative least squares smoothing based on half-quadratic optimization. Due to the presence of PRNU information in the residual between the pre-smoothed and post-smoothed signals, additional cyclic smoothing is applied to refine the signal further. Finally, the smoothed signals are accumulated to obtain the enhanced PRNU noise. Experimental comparisons conducted on the public dataset Dresden demonstrate that the proposed model significantly outperforms existing methods in terms of source camera identification for low-resolution and strong JPEG compression images. [ABSTRACT FROM AUTHOR]

In the context of temporal image forensics, it is not evident that a neural network, trained on images from different time-slots (classes), exploits solely image age related features. Usually, images taken in close temporal proximity (e.g. , belonging to the same age class) share some common content properties. Such content bias can be exploited by a neural network. In this work, a novel approach is proposed that evaluates the influence of image content. This approach is verified using synthetic images (where content bias can be ruled out) with an age signal embedded. Based on the proposed approach, it is shown that a deep learning approach proposed in the context of age classification is most likely highly dependent on the image content. As a possible countermeasure, two different models from the field of image steganalysis, along with three different preprocessing techniques to increase the signal-to-noise ratio (age signal to image content), are evaluated using the proposed method. • A novel approach in the context of XAI and deep learning image age approximation. • The approach evaluates the influence of image content on the classification accuracy. • The approach is verified using synthetic images with an age signal embedded. • Deep learning age approximation can be strongly biased by image content. • Different CNN models trained in the context of age classification are evaluated. [ABSTRACT FROM AUTHOR]

Copyright of Chinese Journal of Intelligent Science & Technology is the property of Beijing Xintong Media Co., Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Copyright of Journal of Signal Processing is the property of Journal of Signal Processing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

From the information forensics point of view, it is important to correctly classify between natural images (outputs of digital cameras) and computer-graphics images (outputs of advanced graphics rendering engines), so as to know the source of the images and the authenticity of the scenes described in the images. It is challenging to achieve good classification performance when the forensic classifier is tested on computer-graphics images generated by unknown rendering engines and when we have a limited number of training samples. In this paper, we propose two simple yet effective methods to improve the classification performance under such challenging situations, respectively based on data augmentation and the combination of local and global prediction results. Compared with existing methods, our methods are conceptually simple and computationally efficient, while achieving satisfying classification accuracy. Experimental results on datasets comprising computer-graphics images generated by four popular and advanced graphics rendering engines demonstrate the effectiveness of the proposed methods.

Convolutional neural networks (CNNs) have achieved impressive successes in fake face image detection. However, CNNs ignore tampering traces outside their attention scope. Moreover, example forgetting events can also pose negative impacts on the face forgery detection accuracy. To address these issues, this paper proposes an attention-expanded two-stage face forgery detector, named Attention-expanded Deepfake Spotter (ADS). In the first stage, the manipulated regions are preliminarily located by utilizing the Region Score Maps (RSMs) generated by the modified CNN. In the second stage, the Expanding and Undetectable Regions (EUR) loss function is designed to encourage another modified CNN to mine manipulation traces outside the manipulated areas exposed in the first stage. To fuse the manipulation traces extracted from different regions in the two stages and mitigate the problems caused by example forgetting events, RSM-weighted accumulation is adopted to integrate the detection information from both stages and obtain the final detection result. The proposed algorithm's effectiveness for each component is analyzed through ablation experiments, and the method is evaluated on four publicly available datasets: FF++, HFF, DFDC, and Celeb-DF. The experimental results demonstrate that the proposed method has high detection rates and superior transferability, outperforming most existing algorithms. [ABSTRACT FROM AUTHOR]

Digital images can be manipulated with recent tools. Image forensics examines the image from several angles to spot any anomalies. Most techniques are applicable to detect a single operation on the image. In actual practice, fake photos are manipulated with multiple operations and compression algorithms. A convolutional neural network with a reasonable size is designed to detect operators and the respective sequences for two operators in particular. The bottleneck strategy is incorporated to optimize the network training cost and a high-depth network. The detection of a particular operator depends on inherent statistical information. A single global average pooling layer preserves the statistical information in a convolutional neural network. The strength of existing detection techniques is also reduced in low-resolution and high-compression environments. The proposed method performs better than existing techniques on compressed small-size images even though forensic is difficult in small-size and compressed images due to inadequate statistical traces. The proposed convolutional neural network also applies to detect operators with unknown specifications and compression not used in training. [ABSTRACT FROM AUTHOR]

The traditional semi-fragile watermarking can effectively protect the integrity of the host image, but the permanent distortion caused by watermark embedding makes the application in high-fidelity images limited. Hence, we propose a novel semi-fragile reversible watermarking scheme for tampering localization. The motivation of this research is to solve the problem that the original carrier cannot be recovered in semi-fragile watermarking method and to improve the accuracy of tampered image detection and location. The proposed scheme embeds the semi-fragile watermark into the mid-frequency quaternion discrete Fourier transform (QDFT) coefficients to provide semi-fragility and reversibly embeds the compensation information into the watermarked image to achieve cover-lossless. First, to reduce the amount of compensation information, the watermarked image is pre-recovered using the neighborhood coefficients of the semi-fragile watermark embedding position. Then, the difference between the pre-recovered image and the original image is represented by the frequency domain error and the spatial domain error. Secondly, the performance of tampering localization is improved by optimizing the graph structure and weight matrix in the manifold ranking (MR) model. Specifically, a new approximate k-regular graph is constituted by introducing spatial relations of neighbors, that must be linked and must not be linked, which are integrated into the original graph structure. Then, a new weight matrix is constructed by combining multi-view features, to enhance the shape performance of the localized regions. The experimental results show that the proposed method can effectively resist content retention attacks, and accurately locate tampered regions. In the event of a non-attack, the original image can be recovered losslessly. [ABSTRACT FROM AUTHOR]

Image recoloring is used to colorize historic grayscale images. Recoloring gives new life to grayscale images with realistic colors. Several techniques exist that can colorize grayscale images. Sometimes attackers create a realistic fake image using colorization techniques. Consequently, there is a need for some technique that can differentiate between natural color and colorized image to stop the misuse of the colorization techniques. In this paper, a Polycolor Model Binary Pattern (PMBP) is proposed to extract robust internal statistical features. Several color models such as RGB, YUV, YCbCr, and HSV are considered to fetch crucial statistical information from an image. A novel Polycolor Model Binary Pattern is formed using the effective channels of several color models. The proposed method gives promising results on non-compressed colorized images as well as on highly compressed colorized images. The efficacy of the proposed detection technique is verified on three fully automated deep learning-based colorization techniques. Three diverse image databases are used to assess the performance. Linear Discriminant Analysis classifier is utilized to classify natural color and colorized images from extracted features, whereas the inverse of the covariance matrix is calculated by using the Moore-Penrose Pseudo Inverse Matrix (MPPM). [ABSTRACT FROM AUTHOR]

Image forensics encompasses a set of scientific tests to investigation of a suspected event via intrinsic clues of imaging pipeline. Traditional image sensing at the Nyquist-Shannon rate as well as the new modality of compressive imaging below the rate are two main types of sensing in photography and imaging applications. Hence, for forensic investigators, it would importantly necessitate the ability to discriminate among images captured by them. However, due to the complex nonlinear nature of imaging processes, investigating imagers' traces is a difficult task. To this intent, we first systematically model the imaging pipelines as an encoder-decoder pair. For exploring distinguishable traces, we mathematically simplify and linearize the pair for compressive imaging and two main forms of traditional image sensing with or without compression. Our theoretical analyses on the approximate linear models reveal blurring kernels of different imagers have discriminability. To validate it in real-world scenarios, we considered the whole imaging process as an inverse problem and estimated the blurring kernel based on a deconvolution approach, where the discriminability is also justified by information visualization. Then, we designed a pipeline classification system, where a deep convolutional neural network is trained by the estimated blurring kernels to be able to classify the three imaging systems. Our results in compressive imaging identification show an accuracy improvement about 3.7 % in comparison to the best result among compared methods. Implementation codes are available for research and development. [ABSTRACT FROM AUTHOR]

In this paper, a novel two-stream convolutional neural network (CNN) method is proposed to determine whether or not the background of an image is replaced with that of another image. Although removing or changing the background of an image is one of the most common image forgery techniques, to the best of our knowledge, this is the first passive image forgery detection (IFD) method in the literature that directly addresses such forgery. The proposed background forgery detection network (BFDNET) consists of two identical DenseNet-based feature extractors and a CNN-based classifier. The method first divides the given image into the foreground containing people in the image and background regions using the Mask R-CNN model in the preprocessing phase. Then, the features from the foreground and background regions are extracted using feature extractors. At last, the feature pair of the regions are combined and classified using a CNN-based classification network to determine if the background of the input image is changed. While conducting this study, the findings of some ablation experiments are used to determine the structure of the model and some important strategies. In addition, the proposed method is compared with nine state-of-the-art image splicing detection (ISD) methods using a novel dataset in the experiments. Experiments demonstrate that the proposed method achieves an accuracy of 95.5% in detecting background forgeries. The proposed method can improve the accuracy by over 14% compared to prior works. [ABSTRACT FROM AUTHOR]

Image sharpening detection, as a crucial branch of image forensics research, has attained a satisfactory level of performance with the assistance of deep learning. However, due to the nature of convolutional neural network (CNN) models, adversarial examples synthesized by generative adversarial networks (GANs) can easily attack existing forensics models. Therefore, deep learning-based forensics faces new challenges. In this paper, a novel architecture inspired by ConvNext is proposed to detect synthesized adversarial USM sharpening images. Through practical demonstration, our proposed technique achieves satisfying performance in recognizing adversarial samples that outperform previous sharpened image forensic systems. In addition, we have undertaken an ablation analysis of our suggested network topology and analyzed the efficacy of different enhancements. [ABSTRACT FROM AUTHOR]

The classical JPEG compression is a rich source of cues for forensic image analysis. However, this compression standard will in the near future be complemented by a new, highly efficient learning-based compression standard called JPEG AI. JPEG AI is fundamentally different from classical JPEG. Hence, its forensic traces can also be expected to be fundamentally different. We argue that there is a pressing need for image forensics research to investigate these traces. In this work, we characterize forensic compression traces of different AI compression algorithms. Our analysis investigates AI compression artifacts in frequency domain and in spatial domain. Both domains exhibit similar artifacts that likely stem from upsampling operations of the decoders. Additionally, we report for one AI codec another artifact in homogeneous regions. We also investigate the artifact detectability in several scenarios including unseen AI compression traces and postprocessing. Here, frequency and autocorrelation features are better on additive noise and classical JPEG post-compression, while RGB features perform better on blurred and downsampled images. [Display omitted] • Characterization of forensic cues of different AI-based image codecs. • Frequency domain and autocorrelation exhibit periodic artifacts. • Artifacts stem likely from upsampling operations and homogeneous regions. • Experiments show that these artifacts are well detectable with deep neural networks. • Further experiments show that detection generalization of artifacts is challenging. [ABSTRACT FROM AUTHOR]

Detection of double Joint photographic experts group (JPEG) compression is a major part of image forensics, especially double JPEG compression detection with the same quantization matrix is still a challenging problem. Detection of double JPEG compression for different quantization matrices has achieved good performance so far, and some methods for detecting the same quantization matrix have been proposed. However, most methods designed for grayscale images, where they utilize only truncation and rounding errors, not fully exploiting all the useful information of color JPEG images. In addition, there is no suitable model to analyze the convergence of JPEG images. These results in poor accuracy of the double compression detection for color JPEG images with the same quantization matrix. To solve this problem, we propose an effective method that utilizes third-order polynomial fitting for error convergence curves. Firstly, five successive compressions are used to extract the quantization error, dequantization error, coefficient error, and truncation error in each compression and decompression process. Secondly, based on the convergence characteristics of these four types of error information, a third-order polynomial model and least squares are applied for the fit, and then, the slope of each point of the fitted curve and the difference of the slopes are used as convergence characteristics. Moreover, the fitted curves predict the pre-compression state, which can effectively amplify the gap between single and double compression. Finally, support vector machines (SVM) are used to train and predict the final features. The experiments illustrate that the proposed method enhances accuracy by 2–3% compared to the most recent method and by approximately 15% compared to the earlier method in multiple datasets, including NRCS, MIXDATA, and MIXDATA-I. [ABSTRACT FROM AUTHOR]

The rapid development of the Generative Adversarial Network (GAN) makes generated face images more and more visually indistinguishable, and the detection performance of previous methods will degrade seriously when the testing samples are out-of-sample datasets or have been post-processed. To address the above problems, we propose a new relational embedding network based on “what to observe” and “where to attend” from a relational perspective for the task of generated face detection. In addition, we designed two attention modules to effectively utilize global and local features. Specifically, the dual-self attention module selectively enhances the representation of local features through both image space and channel dimensions. The cross-correlation attention module computes similarity between images to capture the global information of the output in the image. We conducted extensive experiments to validate our method, and the proposed algorithm can effectively extract the correlations between features and achieve satisfactory generalization and robustness in generating face detection. In addition, we also explored the design of the model structure and the inspection performance on more categories of generated images (not limited to faces). The results show that RENet also has good detection performance on datasets other than faces. [ABSTRACT FROM AUTHOR]

This article introduces the synergistic dual‐stream network (SDS‐Net), a novel neural network architecture that significantly enhances the detection of image manipulations. SDS‐Net employs a unique dual‐stream fusion strategy that processes both RGB image and the corresponding noise map. It innovatively combines features computation from the different blocks of dual backbones, and leverages a multi‐scale spatial pyramid pooling (MSPP) module to expand the receptive fields of shallow features. This approach not only enriches the feature representation, but also ensures the precise localization of manipulated regions. Extensive experiments conducted on various public datasets demonstrate the superiority of SDS‐Net over several state‐of‐the‐art methods. [ABSTRACT FROM AUTHOR]

In this digital era, a huge amount of images are flooding the internet, which is extensively used for digital communications, and are also regarded as a significant source of information in many fields. However, the images can be easily altered without leaving any traces due to the availability of digital image editing softwares. Hence, it becomes essential to validate the integrity of the images. One of the most serious and popular tampering procedures is Copy Move Forgery (CMF), wherein some portion of an image is copied and pasted to another region in the same image. This paper reviews recent state-of-the-art copy-move forgery detection (CMFD) schemes along with their pros, and cons with the help of tables for better readability. In addition, this paper enlists the performance evaluation criteria and different image datasets used for CMFD, along with their merits and demerits. At last, this review addresses the various issues, challenges, and future directions in the field of CMFD. This survey paper aims to provide researchers with a broad perspective on the various aspects of advancements in CMFD techniques. [ABSTRACT FROM AUTHOR]

From the information forensics point of view, it is important to correctly classify between natural images (outputs of digital cameras) and computer-graphics images (outputs of advanced graphics rendering engines), so as to know the source of the images and the authenticity of the scenes described in the images. It is challenging to achieve good classification performance when the forensic classifier is tested on computer-graphics images generated by unknown rendering engines and when we have a limited number of training samples. In this paper, we propose two simple yet effective methods to improve the classification performance under such challenging situations, respectively based on data augmentation and the combination of local and global prediction results. Compared with existing methods, our methods are conceptually simple and computationally efficient, while achieving satisfying classification accuracy. Experimental results on datasets comprising computer-graphics images generated by four popular and advanced graphics rendering engines demonstrate the effectiveness of the proposed methods. [ABSTRACT FROM AUTHOR]

Abstract Morphing attacks (MAs) pose a substantial security threat to the Automatic Border Control (ABC) system. While a few morphing attack detection (MAD) methods have been proposed, the face morphing accomplice's facial restoration has not received sufficient attention. Due to the inability to foresee the morphing factor used for a particular morphed image, selecting the appropriate de‐morphing factor becomes a challenging problem in the restoration of the accomplice's facial image. If the morphing factor cannot be chosen reasonably, achieving the desired restoration effect is difficult. Therefore, this paper presents an adaptive de‐morphing factor framework (ADFF) architecture for restoring the accomplice's facial image. By exploiting the morphed images stored in the electronic passport system and the real‐time captured criminal's images, ADFF can effectively restore the accomplice's facial image. Experimental results and analysis show that ADFF can significantly reduce the security threats of MAs on ABC.

Nowadays, powerful image editing applications not only simplify image processing significantly but also enhance the realism of processed digital images. However, this convenience has presented unprecedented challenges in verifying the authenticity of images. Although existing methods have achieved significant results in image forgery localization, most of them struggle to obtain satisfactory performance when dealing with tampered areas of various sizes, especially for large-scale tampered regions. To enhance the localization performance for various types and sizes of tampered regions, we propose a novel dual-stream intermediate fusion network for image forgery localization, named DIF-Net. This network adopts an encoder-decoder architecture composed of an adaptive convolutional pyramid and dual-stream intermediate fusion modules. Specifically, the former extracts multi-scale information from different depths by utilizing two depth-wise strip convolutions instead of standard large-kernel convolutions. Moreover, during feature fusion, learnable parameters are employed to dynamically allocate weights to each feature scale, so that the network can adaptively select the most relevant features at the target scale. The latter effectively reduces category information differences between the two feature streams by utilizing two learnable intermediate representations to model channel and spatial consistency in the dual-stream features. Compared to traditional and previous deep learning methods, the DIF-Net can generate high-quality prediction masks with fewer parameters. Through extensive experimental validation, our DIF-Net demonstrates outstanding performance on various datasets, surpassing the state-of-the-art forgery localization methods currently available. On the commonly used CASIA2 dataset, our DIF-Net achieves an improvement of 3.3% in F1 and 2.4% in AUC compared to previous methods.

The last years have witnessed significant developments in image acquisition systems and in algorithms for extracting information from them. Nevertheless, in many scenarios, several factors can hinder the recovery of useful data from images. This is especially true and important in forensic applications, where images are often accidentally captured by an imaging system not engineered for that specific acquisition (for example, the surveillance system designed to monitor the entrance of a bank may accidentally capture the license plate of a vehicle passing outside the bank). Therefore, the acquired images often need to be processed to facilitate extracting information from them. When facing a combination of several impairment factors, such as blur and perspective distortion, several image restoration algorithms must be applied. Then, it is necessary to choose a restoration order, which means the order by which single restoration algorithms are chained together to obtain the enhanced image. This study aims to understand whether such an order may impact the final result. Of course, there exists a wide variety of image impairments; in this study, we focus on the case of an image affected by a combination of optical/motion blur, perspective distortion, and additive noise, which are all widespread artifacts in forensic image applications. To answer the question about the importance of choosing one restoration workflow over another, we first model each considered defect and its restoration operator and then analyze and compare the effects of the composition of such operators on the restored output. Such a comparison is made from both a mathematical and experimental point of view, using both images with synthetically generated impairments and pictures with real degradations. The results show that the restoration order can affect significantly the results, especially when the defects are severe.

We propose a novel Tri-Path Backbone Network (TPB-Net) and train it end-to-end to effectively detect multiple types of image manipulations. The key challenge for image manipulation localization lies in the difficulty and diversity of extracting forgery features. To address this, we adopt a Triple-path Interconnected Backbone (TIB) scheme as the feature extractor, which enables the strong feature detection capabilities. Furthermore, we design and introduce the Dual-path Compressed Sensing Attention (DCSA) module, that incorporates a dual-path attention mechanism. The DCSA module intelligently compresses channels in the spatial path and spatial information in the channel path. These compression operations lead to improved learning efficiency, enhanced representation effectiveness, and increased model robustness. TPB-Net offers an end-to-end framework comprising trainable modules, facilitating joint optimization and enabling the achievement of optimal performance. Through rigorous experiments conducted on four standard image manipulation datasets, we demonstrate the superior performance of our method compared to previous state-of-the-art approaches.

Latent fingerprint restoration is one of the most challenging problems for fingerprint identification. According to the current next-generation identification (NGI) system fact sheet, the number of unsolved latent fingerprint cases is more than a million and is increasing daily. The current state-of-the-art algorithms utilize various deep-learning networks to restore latent fingerprints in the spatial domain. Unfortunately, these algorithms faced difficulty recovering genuine friction ridges from low-quality latent fingerprints with complex backgrounds and noise. Instead of using deep-learning models to restore latent fingerprints directly in the spatial domain, we used the deep-learning model to predict a frequency-domain filter for restoring each latent input partitioning block. We integrated this new filter predictor model within a progressive feedback framework. As a result, the proposed method can reveal genuine friction ridges from complex backgrounds, especially around very low-quality areas and singular point areas, which is crucial to the success of latent fingerprint identification. We benchmarked our proposed method against state-of-the-art methods with four public latent fingerprint datasets. In addition, we used one commercial-of-the-shelf (COTS) and another open-source fingerprint matchers in our benchmark. The experiments showed that the proposed method outperformed current state-of-the-art methods in identification accuracy. Compared to the state-of-the-art methods using deep learning in the spatial domain, the proposed method achieved an average rank-1 identification accuracy improvement of 3.75% and 3.19% of four public datasets with the COTS and open-source fingerprint matchers, respectively. Finally, the execution code was made publicly available for non-commercial purposes at https://github.com/skconan/SFP-Progressive-Feedback-Latent-Fingerprint-Restoration.

This research presents a new approach for identifying instances of copy-move forgeries in digital images by utilizing the Multiscale Detector a Neural Network-based method, which serves as an image key-point detector and descriptor. The act of copy-move manipulation involves the replication and subsequent insertion of a specific segment of an image, intending to modify the overall content of the image. The approach we utilize leverages the sophisticated functionalities of Multiscale Detector, a framework that combines key-point detection with descriptor extraction, to accurately detect and localize instances of copy-move manipulation. The effectiveness of our approach is assessed on a range of copy-move forgeries, encompassing instances that have undergone post-processing and geometric transformations. The experimental findings illustrate the resilience of our approach in identifying instances of manipulations over a diverse range of textured images and various alteration approaches. Furthermore, our approach demonstrates strong performance even when subjected to supplementary processing procedures such as brightness modification, color reduction, contrast adjustment, and blurring. Our suggested method has greater performance when compared to the existing manipulation detection approach, as demonstrated through a comparative analysis. In addition, the algorithm we have developed has high computing efficiency, allowing for real-time detection of forgeries. The methodology employed in this study, which builds upon the Multiscale Detector framework, offers a highly effective approach to the detection of copy-move manipulations in digital images.

A critical task in forensics investigation is the recovery of the manipulation history of the image under analysis. To this aim, considering a typical life-cycle of a digital image, the estimation of the misalignment occurred between consecutive JPEG compressions can be considered an useful starting point to localize forgeries and retrieve information about the camera that took the picture through first quantization matrix estimation. In this work, starting from statistics computed from the AC histograms obtained applying a third JPEG compression, an effective and robust deep learning based approach devoted to estimate the aforementioned misalignment has been designed. Finally, to assess the performance of the proposed solution a series of tests has been conducted at varying of patch sizes, quantization matrices, employed datasets, and comparisons with state-of-the-art solutions.

The advancement of deep forgery technology has significantly impacted the credibility of media content, making the detection of deep forgeries crucial for ensuring media security. Although research on deepfake detection methods has been progressively advancing, current approaches predominantly rely on detecting and identifying artifacts. As deep forgery technology continually improves, high-quality synthetic images and those produced through reconstruction methods have become increasingly sophisticated, rendering artifact and trace detection methods somewhat limited. To address this issue, we introduce a deep forgery detection method that integrates deep neural networks with fine-grained artifact features. Our proposed method simulates diverse facial synthesis data by employing facial color conversion, facial frequency domain conversion, and facial mask deformation and blurring. This trains the deepfake detection model to adapt to various synthesis techniques. The classifier model is trained using multiple perturbations of authentic images, with fine-grained artifact features ensuring the stability of the detection process. Our approach achieves superior accuracy and AUC values on the FF++ and WildDeepfake datasets, demonstrating its effectiveness and adaptability in detecting deep forgeries.

Synthetically-generated images are getting increasingly popular. Diffusion models have advanced to the stage where even non-experts can generate photo-realistic images from a simple text prompt. They expand creative horizons but also open a Pandora's box of potential disinformation risks. In this context, the present corpus of synthetic image detection techniques, primarily focusing on older generative models like Generative Adversarial Networks, finds itself ill-equipped to deal with this emerging trend. Recognizing this challenge, we introduce a method specifically designed to detect synthetic images produced by diffusion models. Our approach capitalizes on the inherent frequency artefacts left behind during the diffusion process. Spectral analysis is used to highlight the artefacts in the Fourier transform of a residual image, which are used to distinguish real from fake images. The proposed method can detect diffusion-model-generated images even under mild jpeg compression, and generalizes relatively well to unknown models. By pioneering this novel approach, we aim to fortify forensic methodologies and ignite further research into the detection of AI-generated images.

In this paper, we introduce the novel Multi-Scale Attention Network (MSA-Net) to address the challenge of locating diverse types and sizes of splicing forgery objects. Previous methods neglect crucial characteristics of global dependencies and size, resulting in imprecise localization on background tampering and small target tampering. To overcome this, we integrate a multi-scale self-attention mechanism to capture global dependencies and fully understand the relationships between spliced objects and untampered areas. Our approach involves inserting multi-scale attention modules that combine the position attention and channel attention modules between convolution layers for feature extraction. The position attention module emphasizes spatial interdependencies, capturing relationships between feature positions. Similarly, the channel attention module captures relationships between channel features. This allows for the preservation of intrinsic details while capturing long-range semantic dependencies, which is beneficial to the detection of splicing forgery objects. Meanwhile, by dividing the feature maps into multiple sub-regions or sub-channels, our attention modules can better preserve the details while capturing long-range semantic information dependencies. Experimental results show that the proposed MSA-Net outperforms several state-of-the-art algorithms with an F1-score of 60.5% and an IOU value of 58.8% on the CASIA dataset. [ABSTRACT FROM AUTHOR]

Morphing attacks (MAs) pose a substantial security threat to the Automatic Border Control (ABC) system. While a few morphing attack detection (MAD) methods have been proposed, the face morphing accomplice's facial restoration has not received sufficient attention. Due to the inability to foresee the morphing factor used for a particular morphed image, selecting the appropriate de‐morphing factor becomes a challenging problem in the restoration of the accomplice's facial image. If the morphing factor cannot be chosen reasonably, achieving the desired restoration effect is difficult. Therefore, this paper presents an adaptive de‐morphing factor framework (ADFF) architecture for restoring the accomplice's facial image. By exploiting the morphed images stored in the electronic passport system and the real‐time captured criminal's images, ADFF can effectively restore the accomplice's facial image. Experimental results and analysis show that ADFF can significantly reduce the security threats of MAs on ABC. [ABSTRACT FROM AUTHOR]

Forged images in JPEG format have a ubiquitous presence in today's world due to the ease of availability of image manipulation tools. This paper presents a deep learning-based system that utilizes the inherent relationship between histograms of quantized DCT coefficients and corresponding quantization step sizes to distinguish between original and forged regions in a JPEG image, based on detecting single and double compressed blocks, without fully decompressing the JPEG image. In this direction, we propose a novel combination of raw histograms of the quantized DCT coefficients and corresponding quantization step sizes and use it as input to the counterpart standard CNN architectures designed to handle the proposed input representation for single vs. double JPEG compression detection. The proposed input is shown to have a minimal effect on performance with various standard CNN architectures that are designed to handle the proposed input for the task at hand. Specifically, we have utilized DenseNet to extract the compression-specific artifacts from the proposed input for additional experiments reported in this work. We considered a publicly available dataset generated with a diverse set of 1,120 quantization matrices. Using the proposed input to learn the compression artifacts outperforms the baseline methods for the blocks of sizes 256 × 256 , 128 × 128 , and 64 × 64 . Furthermore, in the case of test blocks compressed with completely different quantization matrices than matrices used in training, the proposed method outperforms the baseline methods. Consequently, improved forgery localization performances are obtained for forged JPEG images. [ABSTRACT FROM AUTHOR]

With the support of deep neural networks, the existing image manipulation detection (IMD) methods can detect manipulated regions within a suspicious image effectively. In general, manipulation operations (e.g., splicing, copy-move, and removal) tend to leave manipulation artifacts in the high-frequency domain of the image, which provides rich clues for locating manipulated regions. Inspired by this phenomenon, in this paper, we propose a High-Frequency Component Enhancement Network, short for HFCE-Net, for image manipulation detection, which aims to fully explore the manipulation artifacts left in the high-frequency domain to improve the localization performance in IMD tasks. Specifically, the HFCE-Net consists of two parallel branches, i.e., the main stream and high-frequency auxiliary branch (HFAB). The HFAB is introduced to fully explore high-frequency artifacts within manipulated images. To achieve this goal, the input image of the HFAB is filtered out of the low-frequency component by the Sobel filter. Furthermore, the HFEB is supervised with the edge information of the manipulated regions. The main stream branch takes the RGB image as input, and aggregates the features learned from the HFAB by the proposed multi-layer fusion (MLF) in a hierarchical manner. We conduct extensive experiments on widely used benchmarks, and the results demonstrate that our HFCE-Net exhibits a strong ability to capture high-frequency information within the manipulated image. Moreover, the proposed HFCE-Net achieves comparable performance (57.3%, 90.9%, and 73.8% F1 on CASIA, NIST, and Coverage datasets) and achieves 1.9%, 9.0%, and 1.5% improvement over the existing method. [ABSTRACT FROM AUTHOR]

With the improvement of digital image display technology, the "secondary imaging" caused by digital cameras is also gradually popularized, and the quality of the recaptured image formed by this imaging is also getting higher and higher, and this kind of high-quality fake image has caused great threat to digital images security. We propose a neural network-based recaptured image identification algorithm and use the difference between two types of images to build the identification algorithm in the frequency domain. The algorithm uses filtering to obtain the feature images which are the high-frequency and low-frequency filtering images, in order to further distinguish the image differences, the direction of the filtered image obtained from high-frequency images, each direction of the filtered image contains high-frequency information at different angles, and the low-frequency image is downsampled. At the same time, the low-frequency image is downsampled to obtain a multi-scale filtered image. The algorithm extracts the features from previous images as the feature values for classification, and finally uses neural networks for classification to obtain the classification results, and these prove that the algorithm presented is able to differentiate the recaptured images effectively in this paper. [ABSTRACT FROM AUTHOR]

Copy‐move forgeries are a conceptual sort of image retouching that involves duplicating a portion of an image and moving it to a different position within the original image, whether by modifying the duplicated part or just moving it altogether. To identify cloned portions that have been reproduced into the same digitized image, the suggested hybrid features in this article combine using the Hessian and Raw patch features on gray‐level images. Using a suggested model that combines patch features built on key points detected by the Hessian detector on gray‐level image, localization of duplicated and pasted portions of the manipulated image were found. After using the combined features in the matching step, the parallelism condition was applied together with the random sample consensus method to eliminate mismatches. Two databases, GRIP and the image manipulation dataset (IMD), were used for the detection and characterization, and the empirical findings show that the approach was successful in achieving an F1 score 100% for the GRIP database. Additionally, with the IMD database, it produced a 92.13% F1 score. The proposed method was also shown to be effective by obtaining high F1 scores in images where noise, JPEG compression, and scaling attacks were applied to make forgery detection difficult. [ABSTRACT FROM AUTHOR]

Text-to-image diffusion models (DM) have posed unprecedented challenges to the authenticity and integrity of digital images, which makes the detection of computer-generated images one of the most important image forensics techniques. However, the detection of images generated by text-to-image diffusion models is rarely reported in the literature. To tackle this issue, we first analyze the acquisition process of DM images. Then, we construct a hybrid neural network based on attention-guided feature extraction (AGFE) and vision transformers (ViTs)-based feature extraction (ViTFE) modules. An attention mechanism is adopted in the AGFE module to capture long-range feature interactions and boost the representation capability. ViTFE module containing sequential MobileNetv2 block (MNV2) and MobileViT blocks are designed to learn global representations. By conducting extensive experiments on different types of generated images, the results demonstrate the effectiveness and robustness of our method in exposing fake images generated by text-to-image diffusion models. • Generating an image will introduce different traces compared with those of real one. • Attention mechanism is capable of capturing long-range feature interactions. • Global traces can provide reliable information for detecting fake images. • Different artifacts lead to performance degradation on other types of images. [ABSTRACT FROM AUTHOR]

This article introduces the synergistic dual‐stream network (SDS‐Net), a novel neural network architecture that significantly enhances the detection of image manipulations. SDS‐Net employs a unique dual‐stream fusion strategy that processes both RGB image and the corresponding noise map. It innovatively combines features computation from the different blocks of dual backbones, and leverages a multi‐scale spatial pyramid pooling (MSPP) module to expand the receptive fields of shallow features. This approach not only enriches the feature representation, but also ensures the precise localization of manipulated regions. Extensive experiments conducted on various public datasets demonstrate the superiority of SDS‐Net over several state‐of‐the‐art methods.