Search

In recent years, with the rapid development of deepfake technology, a large number of deepfake videos have emerged on the Internet, which poses a huge threat to national politics, social stability, and personal privacy. Although many existing deepfake detection methods exhibit excellent performance for known manipulations, their detection capabilities are not strong when faced with unknown manipulations. Therefore, in order to obtain better generalization ability, this paper analyzes global and local inter-frame dynamic inconsistencies from the perspective of spatial and frequency domains, and proposes a Local region Frequency Guided Dynamic Inconsistency Network (LFGDIN). The network includes two parts: Global SpatioTemporal Network (GSTN) and Local Region Frequency Guided Module (LRFGM). The GSTN is responsible for capturing the dynamic information of the entire face, while the LRFGM focuses on extracting the frequency dynamic information of the eyes and mouth. The LRFGM guides the GTSN to concentrate on dynamic inconsistency in some significant local regions through local region alignment, so as to improve the model’s detection performance. Experiments on the three public datasets (FF++, DFDC, and Celeb-DF) show that compared with many recent advanced methods, the proposed method achieves better detection results when detecting deepfake videos of unknown manipulation types.

Classic video compression methods usually suffer from long encode time and requires large memories, making it hard to deploy on edge devices; thus, video compressive sensing, which requires less resources during encoding, is receiving more attention. We propose a robust mixed-rate ROI-aware video compressive sensing algorithm for transmission line surveillance video compression. The proposed method compresses foreground targets and background frames separately and uses reversible neural network to reconstruct original frames. The result on transmission line surveillance video data shows that the proposed compressive sensing method can achieve 26.47, 34.71 PSNR and 0.6839, 0.9320 SSIM higher than existing methods on 1.5% and 15% measurement rates, and the proposed ROI extraction net can precisely retrieve regions under high noise levels. This research not only demonstrates the potential for a more efficient video compression technique in resource-constrained environments, but also lays a foundation for future advancements in video compressive sensing techniques and their applications in various real-time surveillance systems.

Recent studies have revealed that deep neural networks (DNN) used in artificial intelligence systems are highly vulnerable to adversarial sample-based attacks.To address this issue, a dual adversarial attack method was proposed for license plate recognition (LPR) systems in a DNN-based scenario.It was demonstrated that an adversarial patch added to the pattern location of the license plate can render the target detection subsystem of the LPR system unable to detect the license plate class.Additionally, the natural rust and stains were simulated by adding irregular single-connected area random points to the license plate image, which results in the misrecognition of the license plate number.or the license plate research, different shapes of adversarial patches and different colors of adversarial patches are designed as a way to generate adversarial license plates and migrate them to the physical world.Experimental results show that the designed adversarial samples are undetectable by the human eye and can deceive the license plate recognition system, such as EasyPR.The success rate of the attack in the physical world can reach 99%.The study sheds light on the vulnerability of deep learning and the adversarial attack of LPR, and offers a positive contribution toward improving the robustness of license plate recognition models.

Aiming at the problems that “over optimizing” the quality of steganographic text and lack of constraints on the semantic expression of the generated steganographic text in existing generative text steganography methods, a generative text steganography method was proposed based on emotional expression in semantic space.In order to make use of the scene fusion provided by the new media platform to obtain many camouflage scenes, the focus was how to use the unsupervised extraction model to extract the emotional expression combination candidate set from the original data set, then sort the candidate set of emotional expression combinations based on the improved bipartite graph sorting algorithm to obtain the emotional expression combination set, map them to the semantic space, and then implement embedding secret information while generating the user’s opinions based on the emotion expression combinations.Experimental results show that, compared with the existing generative text steganography methods in semantic space, the product reviews generated by the proposed method have a minimum perplexity of 10.536, and have a strong correlation with the chosen product, which can further guarantee the cognitive concealment of steganographic texts.At the same time, the proposed method can also be effectively used in the field of secure and confidential communication, and can avoid the senders being traced and analyzed.

Botnet is one of main threats towards the Internet.Currently, botnets can expand to the whole world due to various types of network services, pervasive security vulnerabilities and massive deployment of networked devices, e.g., internet of things (IoT) devices.Future botnets will become more cross-platform and stealthy, which introduces severe security risks to cyberspace.Therefore, in-depth research on botnets can offer study targets to corresponding defensive studies, which is very meaningful for designing an architecture to secure the next-generation cyberspace.Hence, an HTTP-based scalable botnet framework was proposed to address the problems of compatibility, stealthiness and security.Specifically, the framework adopted a centralized controlling model.Moreover, it used the HTTP protocol as the designed botnet’s communication protocol and block encryption mechanisms based on symmetric cryptography to protect the botnet’s communication contents.Furthermore, a secure control mechanism for multi-platform botnets was designed.In particular, the proposed mechanism utilized source-level code integration and cross-compilation techniques to solve the compatibility challenge.It also introduced encrypted communication with dynamic secret keys to overcome the drawbacks of network traffic regularity and ease of analysis in traditional botnets.Moreover, it designed server migration and reconnection mechanisms to address the weakness of single-point-failure in centralized botnet models.Simulation results in three experimental scenarios with different levels of botnet controllability show that there is a linear relationship between the size of a botnet and the service overhead of the related C＆amp;C servers.In addition, under the condition of the same botnet scale, a higher level of controllability introduces a higher throughput and a greater system overhead.The above results demonstrate the effectiveness and the practical feasibility of the proposed method.

A copyright protection algorithm based on differential privacy for deep fake fingerprint detection model (DFFDM) was proposed, realizing active copyright protection and passive copyright verification of DFFDM without weakening the performance of the original task.In the original task training, noise was added to introduce randomness, and the expected stability of the differential privacy algorithm was used to make classification decisions to reduce the sensitivity to noise.In passive verification, FGSM was used to generate adversarial samples, the decision boundary was fine-adjusted to establish a backdoor, and the mapping was used as an implanted watermark to realize passive verification.To solve the copyright confusion caused by multiple backdoors, a watermark verification framework was designed, which stamped the trigger backdoors and identified the copyright with the help of time order.In active protection, to provide users with hierarchical services, the key neurons in the task were frozen by probabilistic selection strategy, and the access rights were designed to realize the thawing of neurons, so as to obtain the right to use the original task.Experimental results show that the backdoor verification is still effective under different model performance, and the embedded backdoor shows a certain robustness to the model modification.Also, the proposed algorithm can resist not only the collusion attack by the attacker to recruit legitimate users, but also the fine-tuning and compression attacks caused by the model modification.

Aiming at the problems of multiple copies and permanent storage of the transactions containing secret information in the existing storage-based covert communication schemes in blockchain network, as well as the issue of low hidden capacity of the existing time-based schemes, a covert communication method of blockchain network based on transaction construction and forwarding mechanism was proposed.First, an invalid transaction was created by the sender with the help of the transaction construction mechanism, and the secret information was embedded in it, and then the transaction forwarding mechanism was used to send the invalid transaction to the neighbor nodes to form a covert communication channel model, enabling that the transactions containing secret information were spread between nodes and did not exist in the blockchain ledger, so as to achieve the purpose of information concealment and safe transmission.Experimental results show that the transmission capacity of the proposed method is higher than that of the existing schemes, and the single communication time is reduced to 2.5 s.

In order to solve the problems of inaccurate temporal detection and location of the object removal tampered video， a video tamper forensics method based on 3D dual-stream network was proposed.Firstly， the spatial rich model （SRM） layer was used to extract the high-frequency information from video frames.Secondly， the improved 3D convolution （C3D） network was used as the feature extractor of the dual-stream network to extract the high-frequency information and low-frequency information from the high-frequency frame and the original video frame respectively.Finally， through compact bilinear pooling （CBP） layer， two sets of different feature vectors were fused into one set of feature vectors for classification prediction.The experimental results demonstrate that the classification accuracy of the proposed method in all video frames has an advantage in SYSU-OBJFORG dataset， which makes the temporal detection and location of object removal tampered video more accurate.

In order to solve the problem that it is difficult for the existing adversarial disturbance techniques to effectively reduce the discrimination ability of the trackers and make the trajectory deviation rapidly, an effective object tracking adversarial attack method was proposed.First, deception loss, drift loss and attention mechanism-based loss was designed to jointly train generator based on the consideration of the high-level categories and the low-level features.Then, the clean image was sent to the trained generator to generate the adversarial samples that were used to interfere with the object trackers, which made the object trajectory deviation and reduced the tracking accuracy.Experimental results show that the proposed method achieves 54% reduction in success rate and 70% reduction in accuracy on OTB dataset, which can attack the object of tracking quickly in complex scenes.

Aiming at the abnormal behaviors in SDN architecture, a self-protection system based on Renyi entropy that implemented a set of detection, diagnosis and defense method of SDN abnormal behaviors was proposed.The system did not need to introduce the third-party measurement equipment, and directly used the flow table information of OpenFlow switches.Firstly, the abnormal network behavior was detected by calculating the characteristic entropy.Then, the information of the OpenFlow flow table was further analyzed to realize the diagnosis of abnormal behavior.Finally, a blacklist mechanism was established.And the system added the hosts with abnormal behavior to the blacklist and blocked the corresponding abnormal traffic.In order to verify the effectiveness of the system, a prototype was developed on the Floodlight controller.The simulation results on Mininet show that the system can effectively detect, diagnose and defend the abnormal behaviors.The system has low deployment cost, which enhances the security of SDN.

Abstract Steganography is one of the important methods in the field of information hiding, which is the technique of hiding secret data within an ordinary file or message in order to avoid the detection of steganalysis models and human eyes. In recent years, many scholars have applied various deep learning networks to the field of steganalysis to improve the accuracy of detection. The rapid improvement of the accuracy of steganalysis models has caused a huge threat to the security of steganography. In addition, another important factor that limits the security of steganography is capacity. The larger the capacity, the worse and more unnatural the visual quality of carrier images after embedded. Therefore, this paper proposes a steganography model—HIGAN, which constructs the encoding network composed of residual blocks to hide the color secret image into another color image of the same size to output a lower distortion and higher visual quality steganographic image. Moreover, it utilizes the adversarial training between the encoder-decoder network and the steganalysis model to improve the ability to resist the detection of steganalysis models based on deep learning. The experimental results show that our proposed model is achievable and effective. Compared with the previous steganography model for hiding color images based on deep learning, the steganography model in this article could achieve steganographic images with higher visual quality and stronger security.

Mobile edge computing provides low-latency service computing for the Internet of Things (IoT). Considering the computational cost of high-quality image steganography in practical mobile applications, we believe that mobile edge computing could provide real-time service computing for covert communications. As a mainstream approach to convert communication, image steganographic algorithms prefer to hide secret data in well-textured regions in order to reduce the possibility of being detected. Recently, the generative adversarial networks (GAN) has become one of the most popular architectures for image steganography. However, the GAN-based image steganographic algorithms directly conduct the secret data embedding on the entire cover images and do not sufficiently take the regional texture complexity into account, which will compromise the anti-detection ability. To address this issue, we propose a novel image steganographic algorithm on the generated foreground object region with rich textures. More specifically, the foreground object region is generated onto a given cover image by the GAN, and the secret data is embedded in the foreground object region simultaneously during the generation of the region. The experimental results show that the proposed method can resist steganalysis effectively without significant degradation of image quality and achieve real-time processing.

Harmful gas leakage accidents in chemical plants have occurred from time to time. The application of mobile robots to find odor source has become one of the hottest research topics. Compared to traditional robots, unmanned aerial vehicle (UAV) is more flexible and safer. Therefore, using multi-UAV to solve pollution source tracking is a meaningful study. In this paper, an air pollution source tracking algorithm based on artificial potential field and particle swarm optimization is proposed. The particle swarm optimization algorithm combined with artificial potential field method is used to guide the UAVs to track the plume and avoid the collisions among them. At the same time, adaptive inertia weights are used to help improve the convergence and the searchability of particles. We not only evaluated this algorithm in simulation experiments but also designed a multi-UAV pollution source tracking platform for real-world experiments. The experimental results show that the algorithm can accurately find the pollution source in a short time.

Unmanned aerial vehicle technology has made great progress in the past and is widely used in many fields. However, they are unable to meet large-scale and complex missions with a limited energy reserve. Only multiple unmanned aerial vehicles (multi-UAV) work together to better cope with this problem and have been extensively studied. In this paper, a new systematic framework is proposed to solve the problem of multi-UAV collaborative task allocation. It is formulated as a combinatorial optimization problem and solved by the improved clustering algorithm. The purpose is to enable multi-UAV to complete tasks with lower energy consumption. As the number of UAVs rises, it also appears the flight safety issues such as collisions among the UAVs, an improved multi-UAV collision-resistant method based on the improved artificial potential field is proposed. Besides, the UAVs connected with the internet are vulnerable to the various type of network attacks, a method based on the intrusion detection system is proposed to resist the network attack during multi-UAV mission execution. We have also proposed an improved method to improve the accuracy of task allocation further. In addition, an online real-time path planning is proposed to enhance the robustness of multi-UAV to cope with sudden problems. Finally, the numerical simulations and real physical flying experiments showed that the proposed method could provide a viable solution for multi-UAV task allocation; moreover, compared with other task allocation methods, our method has great performance.

With the wide application of the Internet of Things (IoT), the risk of data leakage and theft in IoT is gradually increasing since communication channel is public in data transmission. Thus, the IoT security has become a major problem in information security. Steganography is one of the key methods to solve the problems of personal privacy disclosure and covert communication. In order to make sure secure communications, this paper proposes a novel steganography algorithm based on image-to-image translation by adding steganography module and steganalysis module to CycleGAN, adapting to the covert communication and privacy preserving of the IoT. Steganalysis network is used to improve the anti-detection ability of stego image. Moreover, cycle consistent in CycleGAN can guarantee the quality of the generated image. Through the proposed scheme, the stego image can resist steganalysis by monitors to some extent and remain intact. The experimental results show that this method has a better performance than the state-of-the-art approach.

In cloud computing, data owners host their data on cloud servers, and users (data consumers) can access the data from the cloud servers. This new paradigm of data hosting service also introduces new security challenges that require an independent auditing service to check the integrity of the data in the cloud. Some existing methods for checking the integrity of the data cannot handle this problem efficiently and they cannot deal with the error condition. Thus, a secure and efficient dynamic auditing protocol should reject requests that are made with improper authentication. In addition, an excellent remote data authentication method should be able to collect information for statistical analysis, such as validation results. In this paper, first we design an auditing framework for cloud storage systems and propose an efficient and privacy-preserving auditing protocol. Then, we extend our auditing protocol to support dynamic data operations, which is efficient and has been proven to be secure in the random oracle model. We extended our auditing protocol further to support bidirectional authentication and statistical analysis. In addition, we use a better load distribution strategy, which greatly reduces the computational overhead of the client. Last, we provide an error response scheme, and our experiments show that our solution has good error-handling ability and offers lower overhead expenses for computation and communication than other approaches.

Cloud storage has been recognized as the popular solution to solve the problems of the rising storage costs of IT enterprises for users. However, outsourcing data to the cloud service providers (CSPs) may leak some sensitive privacy information, as the data is out of user’s control. So how to ensure the integrity and privacy of outsourced data has become a big challenge. Encryption and data auditing provide a solution toward the challenge. In this paper, we propose a privacy-preserving and auditing-supporting outsourcing data storage scheme by using encryption and digital watermarking. Logistic map-based chaotic cryptography algorithm is used to preserve the privacy of outsourcing data, which has a fast operation speed and a good effect of encryption. Local histogram shifting digital watermark algorithm is used to protect the data integrity which has high payload and makes the original image restored losslessly if the data is verified to be integrated. Experiments show that our scheme is secure and feasible.

Distributed Computing has achieved tremendous development since cloud computing was proposed in 2006, and played a vital role promoting rapid growth of data collecting and analysis models, e.g., Internet of things, Cyber-Physical Systems, Big Data Analytics, etc. Hadoop has become a data convergence platform for sensor networks. As one of the core components, MapReduce facilitates allocating, processing and mining of collected large-scale data, where speculative execution strategies help solve straggler problems. However, there is still no efficient solution for accurate estimation on execution time of run-time tasks, which can affect task allocation and distribution in MapReduce. In this paper, task execution data have been collected and employed for the estimation. A two-phase regression (TPR) method is proposed to predict the finishing time of each task accurately. Detailed data of each task have drawn interests with detailed analysis report being made. According to the results, the prediction accuracy of concurrent tasks’ execution time can be improved, in particular for some regular jobs.