We introduce a new pre-trainable generic representation for visual-linguistic tasks, called Visual-Linguistic BERT (VL-BERT short). VL-BERT adopts the simple yet powerful Transformer model as backbone, and extends it to take both visual linguistic embedded features input. In it, each element of input is either word from sentence, or region-of-interest (RoI) image. It designed fit most downstream tasks. To better exploit representation, we pre-train on massive-scale Conceptual Captions...

A well defined cost function is crucial to steganography under the scenario of minimizing embedding distortion. In this paper, we present a new for spatial image steganography. The proposed designed by using high-pass filter locate less predictable parts in an image, and then two low-pass filters make low values more clustered. Experiments show that steganographic method with makes changes concentrated texture regions, thus achieves better performance on resisting state-of-the-art...

Histogram shifting (HS) is a useful technique of reversible data hiding (RDH). With HS-based RDH, high capacity and low distortion can be achieved efficiently. In this paper, we revisit the HS present general framework to construct RDH. By proposed framework, one get RDH algorithm by simply designing so-called embedding functions. Moreover, taking specific functions, show that several algorithms reported in literature are special cases construction. addition, two novel efficient also...

Generative adversarial network has shown to effectively generate artificial samples indiscernible from their real counterparts with a united framework of two subnetworks competing against each other. In this letter, we first propose an automatic steganographic distortion learning using generative network, which is composed subnetwork and steganalytic discriminative subnetwork. Via alternately training these oppositional subnetworks, our proposed can automatically learn embedding change...

In traditional steganographic schemes, RGB three channels payloads are assigned equally in a true color image. fact, the security of image steganography relates not only to data-embedding algorithms but also different payload partition. How exploit inter-channel correlations allocate for performance enhancement is still an open issue steganography. this paper, novel channel-dependent partition strategy based on amplifying channel modification probabilities proposed, so as adaptively assign...

The security of image steganography is an important basis for evaluating algorithms. Steganography has recently made great progress in the long-term confrontation with steganalysis. To improve steganography, must have ability to resist detection by steganalysis Traditional embedding-based embeds secret information into content image, which unavoidably leaves a trace modification that can be detected increasingly advanced machine-learning-based concept without embedding (SWE), does not need...

Historically, steganographic schemes were designed in a way to preserve image statistics or steganalytic features. Since most of the state-of-the-art methods employ machine learning (ML) based classifier, it is reasonable consider countering steganalysis by trying fool ML classifiers. However, simply applying perturbations on stego images as adversarial examples may lead failure data extraction and introduce unexpected artefacts detectable other In this paper, we present scheme with novel...

In this paper, we point out that SRM (Spatial-domain Rich Model), the most successful steganalysis framework of digital images possesses a similar architecture to CNN (convolutional neural network). The reasonable expectation is performance well-trained should be comparable or even better than hand-coded SRM. However, without pre-training always get stuck at local plateaus diverge which result in rather poor solutions. order circumvent obstacle, use convolutional auto-encoder procedure. A...

Adoption of deep learning in image steganalysis is still its initial stage. In this paper we propose a generic hybrid deep-learning framework for JPEG incorporating the domain knowledge behind rich steganalytic models. Our proposed involves two main stages. The first stage hand-crafted, corresponding to convolution phase and quantization & truncation second compound neural network containing multiple subnets which model parameters are learned training procedure. We provided experimental...

This article aims to improve deep-learning-based surface defect recognition. Owing the insufficiency of images in practical production lines and high cost labeling, it is difficult obtain a sufficient data set terms diversity quantity. A new generation method called defect-generation adversarial network (SDGAN), which employs generative networks (GANs), proposed generate using large number defect-free from industrial sites. Experiments show that generated by SDGAN have better image quality...

The emergence of powerful image editing software has substantially facilitated digital tampering, leading to many security issues. Hence, it is urgent identify tampered images and localize regions. Although much attention been devoted tampering localization in recent years, still challenging perform practical forensic applications. reasons include the difficulty learning discriminative representations traces lack realistic for training. Since Photoshop widely used practice, this paper...

Relating the embedding cost in a distortion function to statistical detectability is an open vital problem modern steganography. In this paper, we take one step forward by formulating process of assignment into two phases: 1) determining priority profile and 2) specifying cost-value distribution. We analytically show that distribution determines change rate cover elements. Furthermore, when cost-values are specified follow uniform distribution, has linear relation with payload, which rare...

Automatic cost learning for steganography based on deep neural networks is receiving increasing attention. Steganographic methods under such a framework have been shown to achieve better security performance than adopting hand-crafted costs. However, they still exhibit some limitations that prevent full exploitation of their potentiality, including using function-approximated neural-network-based embedding simulator and coarse-grained optimization objective without explicitly pixel-wise...

Recent steganalytic schemes reveal embedding traces in a promising way by using convolutional neural networks (CNNs). However, further improvements, such as exploring complementary data processing operations and wider structures, were not extensively studied so far. In this letter, we design new CNN these aspects order to better capture artifacts. Specifically, on the one hand, propose process information diversely with module called diverse activation module. On other build wide structure...

Image editing techniques enable people to modify the content of an image without leaving visual traces and thus may cause serious security risks. Hence detection localization these forgeries become quite necessary challenging. Furthermore, unlike other tasks with extensive data, there is usually a lack annotated forged images for training due annotation difficulties. In this paper, we propose self-adversarial strategy reliable coarse-to-fine network that utilizes self-attention mechanism...

With the rapid development of face forgery techniques, existing frame-based deepfake video detection methods have fell into a dilemma that may fail when encountering extremely realistic images. To overcome above problem, many approaches attempted to model spatio-temporal inconsistency videos distinguish real and fake videos. However, current works by combining intra-frame inter-frame information, but ignore disturbance caused facial motions would limit further improvement in performance....

The standard local binary pattern (LBP) operator shows its versatility in performing image classification-related tasks, including texture analysis, object recognition, and steganalysis. However, a conventional well-designed scheme utilizing LBP histogram-based features does not have obvious advantage when compared with the well-known steganalytic spatial rich model (SRM). In this paper, we propose an adapted version, called threshold (TLBP), to reveal artifacts caused by data embedding....

Public concerns about deepfake face forgery are continually rising in recent years. Most detection approaches attempt to learn discriminative features between real and fake faces through end-to-end trained deep neural networks. However, the majorities of them suffer from problem poor generalization across different data sources, methods, and/or post-processing operations. In this paper, following simple but effective principle representation learning, i.e., towards learning intra-consistency...

Recently, the success of non-additive steganography has demonstrated that asymmetric distortion can remarkably improve security performance compared with symmetric cost functions. However, most current existing additive steganographic methods are still based on distortion. In this paper, for first time we optimize and propose an A3C (Asynchronous Advantage Actor-Critic) framework, called ReLOAD. ReLOAD is composed actor a critic, where former guides action selection pixel-wise modulation,...