Cloud Computing and Security

Wireless sensor networks (WSNs) own large quantities of small, inexpensive sensor nodes over a region of interest. The sensor nodes are randomly deployed in the region. WSNs may suffer coverage problem due to death of the nodes. It may reduce the coverage quality. In this paper, the network maintenance problem is investigated. We firstly introduce the network health indicator model using the sensing probability model and energy consumption model. And then the holes determination problem is converted into the optimization problem, so we employ the Fuzzy C-Mean method to identify the network holes. Finally, network repair method is proposed to maintenance the network. Simulation results show that the proposed method could effectively improve the lifetime of network and sensing quality.

Crowdsensing has attracted more and more attention in recent years, which can help companies or data demanders to collect large amounts of data efficiently and cheaply. In a crowdsensing system, the sensing tasks are divided into many small sub-tasks that can be easily accomplished by smartphone users, and the companies take advantage of the data collected by all the smartphone users to improve the quality of their services. Efficient task assignment mechanism design is very critical for crowdsensing under some realistic constraints. However, existing studies on task assignment issue are still have many limitations, such as most of them are failed to consider the time budget of smartphone users. Therefore, this work studies the optimal task assignment problem in crowdsensing systems, which can maximize the task completion rate with consideration of the time budget of users. We also prove that the optimal task assignment problem is NP-hard, thus we adopt the linear relaxation and greedy techniques to design a near-optimal crowdsensing task assignment mechanism. We also empirically evaluate our mechanism and show that the proposed task assignment mechanism is efficient.

The authentication between the vehicle and the infrastructure is a vital issue for Vehicular Ad Hoc Networks (VANET), which guarantees to verify the user’s identity and avoids the private information leakage. In this paper, a fast authentication protocol is proposed using the group communication and proactive authentication, in which the authentication is achieved by the symmetric encryption. Therefore, it is more efficient. Moreover, the trade-off between the anonymity and accountability is well.

Underwater acoustic sensor networks (UANs) have emerged as a promising technology which can be applied in many areas such as military and civil in recent years. Among these applications, the communication between the devices is crucial for providing the better service to the users. To facilitate the communication, effective communication approaches need to be developed, however, underwater communication may pose more challenges than terrestrial communication due to the unique characteristics of underwater acoustic channel, such as the high latency and the low bandwidth. In this paper, the context-aware technology is introduced to the design of the communication approach, and the context-aware communication approach for the UAN (CACA-UAN) is proposed which consists of an ontology-based context-aware modeling approach for the UAN, context-aware device association and context-aware communication mechanism to improve the overall performance of the underwater communication system. We have demonstrated that the proposed CACA-UAN can reduce the transmission latency and jitter of communication, and increase the efficiency and reliability of the underwater communication system.

Electricity theft is the main form of Non-Technical Loss (NTL) fraud in the traditional power grid. In Smart Grid, NTL frauds co-occur with various attacks and have more variants. Time-based Pricing (TBP) is an attracting feature of Smart Grid, which supports real-time pricing and billing. However, adversaries could utilize TBP to commit NTL frauds, and we name them as TNTL frauds. Different from NTL frauds which “steal” electricity, TNTL frauds “steal” time. Thus, existing schemes cannot detect them. In this paper, we summarize four attack models of TNTL frauds and analyze various attacks in Smart Grid. We eliminate attacks that do not relate to TNTL frauds and propose countermeasures for those related attacks to prevent TNTL frauds.

Interactions occur across social networks, and modeling interactions in dynamic social networks is a challenging research problem that has broad applications. By combining topology in mathematics with field theory in physics, topology potential, which sets up a virtual field via a topology space to reflect individual activities, local effects and preferential attachments in different interactions, has been proposed to model mutual effects between individuals on social networks. In this paper, we take into consideration not only the information of topology structure and content but also two factors, namely, individual mass and interaction strength. From the perspective of smooth evolution of social networks, we propose a method based on dynamic topology potential, which captures the correlations between different changing snapshots of a social network and can be used to model interactions dynamically, so as to quantify the effects of interactions between individuals on dynamic social networks. Finally, we utilize the dynamic topology potential method for user influence analysis, especially for influential user identification, and the experiment conducted on a real-world data set from AMiner demonstrates the feasibility and effectiveness of our method in terms of a measure for network robustness.

With the increase of network size, there are more and more potential vulnerabilities, which makes it difficult to conduct penetration testing in multihost networks. Attack graph is a useful tool for penetration testing to analyze the relevance of vulnerabilities between hosts and provides a visual view for attack path planning. However, previous works on attack graph generation are inefficient and not applicable to practical penetration testing process. In this paper, we propose an automated vulnerability modeling and verification approach for penetration testing, which generates attack graph efficiently and can be applied to attack process. Petri net is adopted for vulnerability modeling and attack graph synthesis. We implement a prototype system named Automatic Penetration Testing System to verify our method. The system is tested in real networks and the experiment results show the efficiency of our approach.

Bandwidth demand forecasting is the basis and foundation of the power communication network planning. In view of the traditional neural network learning speed is slow, the number of iterations is large, and the local optimal problem, an adaptive extreme learning machine model based on the theory of extreme learning machine and K nearest neighbor theory is proposed to predict the bandwidth of electric power communication. The adaptive metrics of inputs can solve the problems of amplitude changing and trend determination, and reduce the effect of the overfitting of networks. The proposed algorithms are validated using real data of a province in China. The results show that this method is better than the traditional neural network, auto regressive model, self organization model, and single extreme learning machine model. It can be used in electric power communication bandwidth prediction.

Recent studies show that cross technology interference is critically important for emergence of innovative applications in today’s wireless networks. Existing approaches either suffer from the constraint of the need of clean preamble, or requiring technological similarities among cross technology systems.This paper makes the first attempt for tackling this stalemate. We propose Narthil (Narthil is a sword in The Lord of the Rings and The Silmarillion. Narthil was broken in the overthrow of Sauron at the end of the Second Age and was later reforged.), an innovative coexistence system based on imperfect interference management, handling packet detection and symbol synchronization with interfered preambles. Moreover, it recovers the channel state information without the presents of clean preamble.In Narthil, simple Butterworth bandstop filter is applied for an imperfect interference filtering. And the residual signals could be used for packet detection, symbol synchronization, and CSI estimation. The insight is, the inherent properties of preamble such as periodicity and modulation scheme, and the continuity in OFDM band could be effectively leveraged for some important operation such as signal detection, synchronization, and CSI estimation.This inspiring design policy could be further leveraged for other cross technology signals. We implement Natrtil on our USRP/GNU radio platform, and evaluate its performance by using 15 USRP N210 devices. The experimental results demonstrates that Narthil could effectively perform the signal detection and synchronization, as well as CSI estimation without the presents of clean preamble.

Wireless communications and networking technologies are the foundations for road safety applications in Vehicular Ad hoc Networks (VANETs). Since VANET employing IEEE 802.11p only suffers from broadcast storm problems at a high vehicle density, many clustering schemes have been proposed and yet still can’t effectively address the interference problems. Later, some scholars envisioned applying Visible Light Communications (VLC) as the wireless communication technology in VANET. However, VLC requires a strict line-of-sight transmission to maintain stable system performance. In this paper, we propose a hybrid architecture which integrates IEEE 802.11p based VANET with the VLC system. Then, we design a novel Multi-hop Clustering Scheme based on the weighed Virtual Distance Detection (MCSVDD) for the safety message delivery. Through network simulations, we demonstrate far superior performance of the IEEE 802.11p-VLC hybrid VANET architecture compared to that of NHop scheme through key metrics such as maximum delay and normalized goodput.

Using meteorological wireless sensor network to monitor the air temperature (AT) can greatly reduce the costs of monitoring. And it has the characteristics of easy deployment and high mobility. But low cost sensor is easily affected by external environment, often lead to inaccurate measurements. Previous research has shown that there is a close relationship between AT and solar radiation (SR). Therefore, We designed a back propagation (BP) neural network model using SR as the input parameter to establish the relationship between SR and AT error (ATE) with all the data in May. Then we used the trained BP model to correct the errors in other months. We evaluated the performance on the data sets in previous research and then compare the maximum absolute error, mean absolute Error and standard deviation respectively. The experimental results show that our method achieves competitive performance. It proves that BP neural network is very suitable for solving this problem due to its powerful functions of non-linear fitting.

To energy-efficient communication, Compressive Sensing (CS) has been employed gradually. This paper proposes a data gathering scheme based on CS. The network is divided into several blocks and each block sends data to the sink for reconstruction. Experiments demonstrate that our algorithm is feasible and outperforms other schemes.

Mobile crowdsensing is a new paradigm that tries to collect a vast amount of data with the rich set of sensors on pervasive mobile devices. However, the unpredictable intention and various capabilities of device owners expose the application to potential dishonest and malicious contributions, bringing forth the important issues of data credibility assurance. Existed works generally attempt to increase data confidence level with the guide of reputation, which is very likely to be unavailable in reality. In this work, we propose CLOR, a general scheme to ensure data credibility for typical mobile crowdsensing application without requiring reputation knowledge. By integrating data clustering with logical reasoning, CLOR is able to formally separate false and normal data, make credibility assessment, and filter out the false ingredient. Simulation results show that improved data credibility can be achieved effectively with our scheme.

In recent years, the auction has been widely applied in wireless communications for spectrum allocation. In this paper, we investigate the online spectrum double auction problem and propose a Profit concerning and Truthful Online spectrum double auction Mechanism (PTOM). Different from most previous works, we consider the dynamic arrival of primary users (PUs) and secondary users (SUs) and allow SUs to request distinct time slots for using the spectrum. By introducing the priority bid, we capture the online and location associated feature to improve the spectrum utility. Based on the priority bid, we design an efficient admission and pricing rule to improve the auctioneer’s profit. Theoretical analyses are provided to prove that our mechanism has nice economic properties including individual rationality, budget-balance and resistance of time-based and value-based cheating.

Smart Grid employs Advanced Metering Infrastructure (AMI) to automatically manage metering and billing processes supporting various advanced features. Electricity theft is not limited to the traditional methods, such as bypassing power lines. In Smart Grid, adversaries could gain illegal benefit from the utility via various new ways, and we call this class of behavior as Non-Technical Loss (NTL) fraud. In this paper, we study various security issues in AMI and figure out which issues could be utilized by NTL fraud. We analyze various attacks in AMI and generalize attack tree of NTL fraud. We propose countermeasures for security issues to prevent NTL frauds.

In order to improve energy efficiency, Compressive Sensing has been employed gradually in the process of gathering data and transmitting information of sensors. In this paper, a mixed idea has been proposed based on classification for actual environments. At its heart lies a simple yet effective thought that the number of transmission of bottom sensors by no-CS schemes is less than ones by CS. In experiments, our scheme has been proved valuable and feasible.

Energy consumption has always been a challenging issue in wireless sensor networks (WSNs). In this paper, we consider the collaboration optimization problem for load balancing with mobility-assisted features. In particular, we present a cluster-based network structure, in which sensor nodes are partitioned into layers according to transmission radius. Based on a distributed scheme for clustering and cluster heads, rendezvous points (RPs) are introduced and searched through greedy algorithm with geometrical relationship between a specific cluster head and its members. After that, mobile sinks are then introduced to replace the cluster heads in the place where RP has been found. Furthermore, mean squared error of energy in a cluster is used to reduce transmitted packets. Considering another factor, i.e. the data packet, of energy consumption model, we propose an energy optimal algorithm through the quantization approach to balance network load and allocate node’s transmitted traffic. Finally, we analyze the cluster lifetimes in different scenarios and achieve balance to save significant energy.

Wireless Sensor Networks (WSNs) have been increasingly deployed in harsh environments for special applications such as ecological monitoring and Volcano monitoring. Harsh environments can easily cause the network to be unconnected, which lead to unable to collect data by multi-hops. The development of the Unmanned Aerial Vehicles (UAVs) makes it possible to collect data from ground sensor nodes by UAVs, and the UAV trajectory planning is necessary for energy conservation and data collection efficiency. The challenge of the trajectory planning is keeping the trajectory as short as possible while ensures the communication constraints to be satisfied. In this paper, we formulate the optimal UAV trajectory planning problem to a mixed integer programming (MIP) problem, and develop a heuristic algorithm to find a feasible solution to this problem. There are four steps of our scheme: initialization, rotation, optimization and smooth. The simulation results show that our trajectory planning scheme can shorten the length of the UAV’s trajectory while satisfy communication constraints for every sensor nodes.

Existing somatosensory games mainly use vision-based methods, which are affected easily by Line-of-Sight (LOS), occlusions, complex background, wearable devices, and so on. To address these limitations, we propose Wi-Play, a robust human activity recognition system using Wi-Fi signals. Based on the intuition that a specified activity introduces a unique pattern in the time-series of waveform for Channel State Information (CSI-waveform), Wi-Play leverages the unique pattern in the CSI-waveform values as the indicator of human activities. Wi-Play consists of two commercial Off-The-Shelf (COTS) WiFi devices, a TP-LINK TL-WR842N router as the transmitter and an Intel NUC D54250WYKH laptop as the receiver. As our major contributions, we propose a human activity detection algorithm for the extraction of CSI-waveform, build multi-classifiers to recognize human activities, and implement a real-time human activity recognition system using a series of novel technologies. Wi-Play achieves more than 88.3 % average recognition accuracy for recognizing human activities.

A smart toy is defined as a device consisting of a physical toy component that connects to one or more toy computing services to facilitate gameplay in the Cloud through networking and sensory technologies to enhance the functionality of a traditional toy. A smart toy in this context can be effectively considered an Internet of Things (IoT) with Artificial Intelligence (AI) which can provide Augmented Reality (AR) experiences to users. Referring to the direction of the United States Federal Trade Commission Children’s Online Privacy Protection Act (COPPA) and the European Union Data Protection Directive (EUDPD), this study adopts the definition of a child to be an individual under the age of 13 years old. In this study, the first assumption is that children do not understand the concept of privacy. The second assumption is that children will disclose as much information to smart toys as they can trust. Breaches of privacy can result in physical safety of child user, e.g., child predators. While the parents/legal guardians of a child strive to ensure their child’s physical and online safety and privacy, there is no common approach for these parents/guardians to study the information flow between their child and the smart toys they interact with. This paper discusses related privacy requirements for smart toys in a toy computing environment with a case study on a commercial smart toy called Hello Barbie from Mattel.

In a traditional wireless sensor network with static sinks, sensor nodes close to the sink run out of their batteries quicker than other nodes due to the increased data traffic towards the sink. These nodes with huge data traffic are easy to become hotspots. Therefore, such networks may prematurely collapse since the sink is unreachable for other remote nodes. To mitigate this problem, sink mobility is proposed, which provides load-balanced data delivery and uniform energy dissipation by shifting the hotspots. However, the latest location update of the mobile sink within the network introduces a high communication overhead. In this paper, we propose Grid Routing, an energy-efficient mobile sink routing protocol, which aims to decrease the advertisement overhead of the sink’s position and balance local energy dissipation in a non-uniform network. Simulation results indicate that Grid Routing shows better performance in network lifetime when compared with existing work.

Most people do not have a consciousness of energy saving. For this phenomenon, the governments are building smart grids to take measures for the energy crisis. Electric load data records the electric consumption and plays an important role in operation and planning of the power system. However, in home, electric load data usually has the abnormal, noisy and missing data due to various factors. With wrong data, we can not analysis the data correctly, then can not take the right actions to avoid the energy wastes. In this paper, we propose a new solution for the electric load data verification and repair in home environment. As the result shows, proposed method have a better performance than the up to date methods.

With the increasing of energy demand and electricity price, researchers gain more and more interest among the residential load monitoring. In order to feed back the individual appliance’s energy consumption instead of the whole-house energy consumption, Non-Intrusive Load Monitoring (NILM) is a good choice for residents to respond the time-of-use price and achieve electricity saving. In this paper, we discuss the system framework of NILM and analyze the challenges in every module. Besides, we study and compare the public data sets and recent approaches to non-intrusive load monitoring techniques.

The success of the Global Positioning System (GPS) makes the demand for location-based services increase rapidly. However, in the indoor environment, since the reception of the satellite signal is disrupted severely, the accuracy of GPS positioning can not meet the requirements. Radio Frequency Identification (RFID) technology with the advantages of non-contact, non-visibility, low cost and high positioning accuracy, begins to get more and more attention and becomes the most suitable indoor positioning technology. RFID location is a technique which is based on signal strength positioning, using the received signal strength indication (Received Signal Strength Indicator, RSSI) to determine the position of the object. In recent years, the indoor positioning technology has made great progress, especially on the verity of the localization algorithm. In this paper, we will briefly describe the basic principles of RFID, then we introduce the algorithms of existing indoor RFID positioning system. Finally, we analyze their strengths and weaknesses.

The localization technology takes an important role in real-time alarm and rapid response. In this paper, we propose a novel localization method for wireless sensor network which is based on Radio Frequency Identification technology (RBLOCA). This method combines above both techniques and achieves the localization in WSN. The method not only has the advantages of real-time monitoring and low cost of WSN, but also has the characteristics of fast and repeatable use of RFID technology. Experimental results show that the proposed method can achieve a higher positioning accuracy. At the same time, the energy consumption of nodes is also very low. Therefore, we can get the method has a wide range of applications in a number of scenarios such as the forest fire alarm system, underwater sensor network.

Nowadays multimedia applications surround our life. The relevant technologies of multimedia have been widely used anywhere, e.g. education, medicine, research, transportation and recreation, as long as people who has a mobile device. The computing ability of mobile device is growing rapidly, but it still has its limitations. One of significant influence on most mobile devices is the active time of them depend on the capacity of battery. Multimedia may make battery runs out quickly because it usually carries a large amount of information and data so that it leads to a great quantity of calculation and bandwidth requirement. In order to alleviate the computation amount of mobile devices, we propose an adaptive cloud-based multimedia conversion system to eliminate energy consumption and unnecessary occupancy on bandwidth in accordance with the capability and network state of users’ mobile devices. Then we propose a Simulated Annealing-based algorithm to solve the optimization problem of scheduling for a multi-user environment.

Color information is important to image quality assessment (IQA). However, most image quality assessment methods transform color image into gray scale, which fail to consider color information. In recent years, color image processing by using the algebra of quaternions has been attracting tremendous attention. Extensive moments based on quaternion have been introduced to deal with the red, green and blue channels of color images in a holistic manner, which have been proved more effective in color processing. With these inspirations, this paper presents a full-reference color image quality assessment metric based on Quaternion Tchebichef Moments (QTMs). QTMs are first employed to measure color and structure distortions simultaneously. Considering that moments are insensitive to weak distortions in high-quality images, gradient is incorporated as a complementary feature. Luminance is also considered as an auxiliary feature. Finally, a QTM-feature-based weighting map is proposed to conduct the pooling, producing an overall quality score. The experimental results on five public image quality databases demonstrate that the proposed method outperforms the state-of-the-arts.

This paper presents a fast algorithm for the aggregation and analysis of ECG data. The whole process of fusion and analysis can be divided into three stages. ECG signal de-noising is the first stage. A combined filter is used to cut out the noises from ECG signals. In the second stage, a simple method named SDTW (the Sample Dynamic Time Wrapping) is proposed to improve the time efficiency of DTW. Then SDTW and K-means algorithm are applied to attain templates as well as compress templates. The last stage is to train a BP neural network with the compressed templates and other ECG features. Experiments with the MIT-BIH arrhythmia database shows that our algorithm can efficiently improve the recognition accuracy and shorten the recognition time.

In this paper, we propose a novel method for near-duplicate image elimination by tracing the original image of each near-duplicate image cluster. To generate a similarity matrix of each cluster, both global feature and local feature are extracted to accurately evaluate the visual similarity of each image pair. According to the similarity matrix, an Image Relational Graph (IRG) is constructed. Then we adopt the graph model based link analysis algorithm PageRank to analyze the contextual relationship between images on this IRG. In this way, the original image will be correctly traced with the highest rank, while other redundant near-duplicate images in the cluster will be eliminated. To validate the performance of our proposed method, large amount of near-duplicate images mixing with distracting images are applied for experiments, and the experimental results indicate the effectiveness of our method.

For the purpose of public security, car detection and identification are urgently required in the real time traffic monitoring system. However, fine-grained recognition is a challenging task in the area of computer vision due to the subtle inter-class and huge intra-class differences. To tackle this task, this paper provided a novel approach focussed on two main aspects. On the one hand, the most discriminative local feature representations of regions of interests (ROIs) magnified many details. On the other hand, the hierarchical relations within the fine-grained categories can be simulated by probability formulas. Our proposed model consists of two modules: (i) a region proposal network to generate plenty of ROIs and (ii) a joint CNN-based model to learn the multi-grained feature representations simultaneously.The proposed joint CNN-based model was implemented and tested on the Stanford Cars dataset and the CompCars dataset. Our experimental results are compared with those of other methods, and verify the superior performance of the proposed model.

We presented in this paper a usage scenario in which cultural resources in a public context, items on display in a historical museum for instance, should be recommended to groups of visitors in response to their interest (or preferences), thus conserving computational resources and reducing network traffic. Motivated by the scenario, we set out to design and implement a group recommender system, Museum Guides for Groups (MGG), that provides visitors to a museum with a sequence of items of interest by efficiently clustering visitors of similar user profiles into groups and computing recommendations for each group. Our work in progress was reported, focusing on the system design and the selection of an appropriate clustering algorithm for dividing visitors. We evaluated the efficiency of three candidate clustering techniques, including the bisecting K-Means, DBSCAN, and improved CURE, using the MovieLens dataset with 1M ratings.

This paper focuses on human facial expression recognition in video sequences. Different from the methods of two-dimensional image recognition and three-dimensional spatial-temporal interest point detection, our approach highlights human facial expression recognition in complex spatial-temporal video datasets. The major challenge in facial expression recognition is how to obtain a feature dictionary from extracted cube pixel windows based on clustering algorithm. In this paper, our contributions are mainly concentrated on two aspects. Firstly, we combine discrete linear filter with key parameters selection procedure to extract 3D cuboids. Secondly, we propose a novel seed spot selection method to optimize K-means clustering algorithm. The proposed algorithms are evaluated on open databases. The results show that our approach can achieve outstanding results and the proposed approach is significantly effective.

Palmprint is an essential biometrics for personal identification, especially in forensic security. Due to the large valid region of palmprint, palmprint recognition is very time-consuming. Fast and accurate palmprint recognition is an urge problem. To speed up matching, some methods based on classification and indexing are proposed, in which accuracy may be dropped. Ridge distance is often used as an auxiliary feature in palmprint recognition. In this paper, a novel minutiae matching algorithm incorporating ridge distance is proposed. Firstly, we incorporate ridge distance to conventional minutiae. Then, with the restriction of ridge distance, palmprint alignment could be greatly sped up. Experimental results show that our method could not only reduce time cost for matching, but also improve matching accuracy.

Anomaly detection for helicopter rotor provides fault early warning and failure detection to avoid catastrophic accidents and major downtime. It is difficult to extract effective fault features from non-stationary and non-linear vibration data of rotor. A novel time-frequency feature is presented based on short-time Fourier transform in the paper. Due to lack of abundant fault data in practice, support vector data description is also exploited to detect damages by building a model only with normal data. We experimentally evaluate the performance of the proposed anomaly detection on realistic vibration data of helicopter rotor. The results demonstrate that the time-frequency features are closely related to the states of rotor, and the anomaly detection algorithm can clearly detect damages.

Quantum steganography is an important branch of quantum information hiding. It integrates quantum secure communication technology and classical steganography, which embeds the secret information into public channel for covert communication. In this paper, based on the novel enhanced quantum representation (NEQR), a new quantum steganography algorithm is proposed to embed the secret information into the quantum carrier image and the quantum watermark image. The second least significant qubit (LSQb) of quantum carrier image is replaced with the secret information by implementing quantum circuit for good imperceptibility. Compared with the previous quantum steganography algorithms, the key shared by communicating parties can recover the secret information even if it was tampered while the tampers can be located effectively as well. In the experiment result, the Peak Signal-to-Noise Ratio (PSNR) is calculated with different quantum carrier images and quantum watermark images, which demonstrates the imperceptibility is good.

For copy-move forgery, the copied region may be rotated or flipped to fit the scene better. A blind image forensics approach is proposed for copy-move forgery detection using rotation invariant uniform local binary patterns ($$LBP_{P, R}^{riu2}$$LBPP,Rriu2). The image is first filtered and divided into overlapped blocks with fixed size. The features are extracted from each block using $$LBP_{P, R}^{riu2}$$LBPP,Rriu2. Then, the feature vectors are sorted and block pairs are identified by estimating the Euclidean distances of these feature vectors. Specifically, a shift-vector counter C is exploited to detect and locate tampering region. Experimental results show that the proposed approach can deal with multiple copy-move forgeries, and is robust to JPEG compression, noise, blurring region rotation and flipping.

Digital images can be easily manipulated due to availability of powerful image processing software. Passive authentication as a common challenge method of digital image authentication is extensively used to detect the copy-move forgery images. In this paper, a passive authentication scheme is proposed to authenticate copy-move forgery based on discrete cosine transform (DCT). At the feature extraction step, DCT is applied to image blocks and makes use of the means of DCT coefficients to represent image blocks. The size of feature vectors are optimized. At the matching step, a set number of packages is used to store the feature vectors. The similar blocks can be found by comparing the feature vectors that are contained in adjacent packages. The experimental results demonstrate that the proposed scheme can locate irregular and meaningful tampered regions and multiply duplicated regions. In addition, it can also locate the duplicated regions in digital images that are distorted by adding white Gaussian noise, Gaussian blurring and their mixed operations.

MP3 is one of the common formats in the recording equipments. The authenticity and integrity of MP3 audio is widely concerned. We analyzed quantization characteristic of MP3 encoding, and studied the effect of quantified in frame offset and non offset frame. Then we combined statistical characteristics of zero spectral coefficients before and after quantization. Finally, a tampering detection method based on quantitative characteristics is proposed. According to the disadvantage of existing frame offset method that cannot detect high rate compression, this paper in the quantization artifacts description according to the characteristics of the line frequency distribution used in front of the 16 band. We can further effectively solve tamper detection problem of high bit rate compression by studying 16 band quantization. The experimental results show that the accuracy rate of proposed detection method can reach up to 99 %, the compression rate of detection can reach 256 Kbps, and the complexity of compared with the existing methods is significantly reduced.

The detection rates of existing universal blind detection reduced greatly in practical applications due to the generalization problem. According to the principle of orthogonal design, this paper builds three sample sets of embedding rates mismatch, embedding algorithms mismatch and image sources mismatch between the training sample and the testing sample. The three sets are used to test the detection error rates of Rich Model in the case of embedding rates mismatch, embedding algorithm mismatch and image source mismatch. This paper proposes several methods to improve the generalization ability of the universal blind detection, including training the sample by small embedding rates, learning various kinds of embedding algorithms, pre-classifying the testing sample and improving the IQM algorithm. The results show that the practicability of the universal blind detection will be improved.

Image manipulation has become commonplace with growing easy access to sophisticated photo editing softwares. One of the most common types of image forgeries is the copy–move forgery, wherein a region from an image is replaced with another region from the same image. Many existing forensic methods suffer from their inability to detect the cloned area, which is subjected to various transformations such as scaling, rotation, flipping and blurring. In this paper, we propose a novel forensic method based on high-speed corner detection (HSCD) technique and improved scale invariant features transform (SIFT) descriptor. Machine learning technique is used to detect feature points which greatly decreasing the processing time compare to other feather detectors. Experimental results show the efficacy of this technique in detecting copy-move forgeries and estimating the geometric transformation parameters. Compared with the state of the art, our approach obtains a higher true positive rate and a lower false positive rate.

The analysis of topic evolution mainly refers to the mining of topic content which evolves as the time goes on. With the assumption that topic content may be embodied by key words, this article adopted word2vec for the training of 750,000 pieces of news and micro-blog texts and thus established the model of word vector. Then, the text information flow was applied into the model and all word vectors by time series were acquired. Finally, the word vectors were clustered by K-means before the key words were drawn and the analysis of topic evolution was visualized. By comparing the effect of the model of word vector on drawing topic with those of LDA or PLSA topic models, the results showed that the former is superior to the latter two models. Besides, to collect abundant and varied data will facilitate the training of the model of word vector with better generalization ability and the investigation on real-time analysis of topic evolution.

In order to protect data privacy, image with sensitive or private information needs to be encrypted before being outsourced to the cloud. However, this lead to difficulties in image retrieval. A secure JPEG image retrieval method is proposed in this paper. Image is encrypted on DCT (Discrete Cosine Transform) domain by scrambling encryption, and the encrypted image is outsourced to the server, then the DC difference histogram and LBP (Local Binary Patterns) among the image blocks are extracted as image feature vectors by the server. Both the image confidentiality and retrieval accuracy are guaranteed in the proposed method with less computational complexity and communication cost. Experimental results prove that the proposed scheme has good encryption security and can achieve better retrieval performance.

In this paper, we propose a method for image authentication using photon counting and double random phase encryption technique in the Fresnel domain. Recently, it was reported that a better avalanche effect can be achieved for double random phase encoding (DRPE) in the Fresnel domain than that in the Fourier domain in bit units. Moreover, it was verified that photon counting technique can enhance the security of DRPE algorithm. Therefore, DRPE in the Fresnel domain combining with photon counting scheme is much safer for image authentication. In this study, an image is first encrypted by DRPE in the Fresnel domain. Then, a photon-limited encrypted image is produced by using photon counting method. Finally, the decrypted image which can not be easily visualized and recognized under a limited number of photon is authenticated with a statistical nonlinear correlation algorithm. Experimental results verify the feasibility of the proposed method.

The nonlinear conjugate gradient (CG) algorithm is one of the most effective line search algorithms for optimization problems due to its simplicity and low memory requirements, particularly for large-scale problems. However, the results of the new conjugacy conditions are very limited. In this paper, we will propose a new conjugacy condition and two CG formulas. Global convergence is achieved for these algorithms, and numerical results are reported for Benchmark problems.

The wheel wear is a dynamic phenomenon that varies with many mechanical and geometrical factors. Accurately estimating wheel wear is a vital issue in wheel maintance. This paper presents a nature-inspired metaheuristic regression method for precisely predicting wheel status that combines least squares support vector machine (LS-SVM) with a novel PSO-GA-LM algorithm. The PSO-GA-LM algorithm integrates Particle Swarm Optimization (PSO), Genetic Algorithm (GA) and Logistic Map (LM). The method is used to optimize the hyper-parameters of the LS-SVM model. The proposed model was constructed with datasets of the tread wear derived from Taiyuan North Locomotive Depot. Analytical results show that the novel optimized prediction model is superior to others in predicting tread wear with lower RMSE (0.037MPa), MAE (0.027MPa) and MAPE (0.0008 %).

The search behavior of firefly algorithm (FA) is determined by the attractions among fireflies. In the standard FA and its most modifications, worse fireflies can move toward other better ones, while better fireflies seldom move to other positions. To enhance the search of better fireflies, this paper presents a hybrid firefly algorithm (HFA), Which employs a local search operator inspired by differential evolution (DE). Moreover, the control parameters are dynamically adjusted during the search process. Experiments are conducted on thirteen continuous optimization problems. Computational results show that HFA achieves better solutions than the standard FA and three other improved FA variants.

In this paper, we introduce a novel pbest updating strategy to improve the achievement of the original particle swarm algorithm (PSO). First, we set a threshold for using our proposed updating strategy for pbest. Then if the algorithm reaches the condition of using this threshold, we select a pbest with an excellent performance in the population to search in a local valuable region for improving the precise search of particles. Meanwhile, we also select a pbest with a worse performance to search in the entire solution space for improving the global search ability of particles. By comparing with the traditional PSO and its variants on benchmark functions, the PSO algorithm with a novel pbest updating strategy (PPSO) performs much better than the other compared algorithms.

Manifold learning is developed to find the observed data’s low dimension embeddings in high dimensional data space. As a type of effective nonlinear dimension reduction method, it has been widely applied to data mining, pattern recognition and machine learning. However, most existing manifold learning algorithms work in a “batch” mode and cannot effectively process data collected sequentially (or data streams). In order to explore the intrinsic low dimensional manifold structures in data streams on-line or incrementally, in this paper we propose a new manifold Learning algorithm based on Incremental Tangent Space Alignment, LITSA for short. By constructing data points’ local tangent spaces to preserve local coordinates incrementally, we can accurately obtain the low dimensional global coordinates. Experiments on both synthetic and real datasets show that the proposed algorithm can achieve a more accurate low-dimensional representation of the data than state-of-the-art incremental algorithms.

In this paper, we introduce a new mining algorithm to improve the classification accuracy rates aiming at the deficiency of the typical K-Dependence Bayes (KDB) model which ignores the topology changing as the result of inputing of test instances. Under this condition, we put forward an algorithm called Base K-Dependence Bayes (B-KDB) which consists of a Label-based K-Dependence Bayes (L-KDB) algorithm and a Instance-based K-Dependence Bayes (I-KDB) algorithm. The I-KDB algorithm is used to build I-KDB model by instances to be tested and it can deal with the problem of test instances topology keep on changing. However I-KDB model is extraordinarily sensitive to data and it may suffer from overfitting and the effect of noisy instances, therefore L-KDB algorithm is designed as complement. After combing these two algorithms into B-KDB, we built B-KDB model and tested the performance against the KDB model in classification accuracy, precision, sensitivity-specificity analysis with 10-fold cross validation on 55 real benchmark datasets from University of California Irvine (UCI) machine learning repository. The experimental result, which shows the classification accuracy of our model twice as much as KDB, indicates our algorithm efficient and proves our idea of improving the KDB algorithm classification accuracy feasible.

Kernelized Correlation Filter (KCF) is one of state-of-the-art trackers. However, KCF suffers from the drifting problem due to inaccurate localization caused by the scale variation and wrong candidate selection. In this paper, we propose a new method, named Scale Invariant KCF (SIKCF), which estimates an accurate scale and models the distribution of correlation response to address the template drifting problem. The features of SIKCF consist in: (1) A scale estimation method is used to find an accurate candidate. (2) The correlation response of the target image is reasonably considered to follow a Gaussian distribution, which is used to select the better candidate in tracking procedure. Extensive experiments on the commonly used tracking benchmark show that the proposed method significantly improves the performance of KCF, and achieves a better performance than state-of-the-art trackers.

Selective ensemble learning has drawn high attention for improving the diversity of the ensemble learning. However, the performance is limited by the conflicts and redundancies among its child classifiers. In order to solve these problems, we put forward a novel method called KNIA. The method mainly makes use of K-means algorithm, which is used in the integration algorithm as an effective measure to choose the representative classifiers. Then, negative correlation theory is used to select the diversity of classifiers derived from the representative classifiers. Compared with the classical selective learning, our algorithm which is inverse growth process can improve the generalization ability in the condition of ensuring the accuracy. The extensive experiments demonstrate that the robustness and precision of the proposed method outperforms four classical algorithms from multiple UCI data sets.

With the development of Internet of Things, aircraft noise monitoring system can be more accurate and real-time by laying large-scale monitoring devices on monitoring areas. In this paper, we present a real-time aircraft noise detection algorithm based on large-scale noise data. The spatial characteristics of the distribution of noises are discussed firstly as the premise of analyzing the differences of aircraft noises and other kinds of noises. Then we propose a way to represent the tendency surface of noise propagation and attenuation, and the unit tendency increment in one direction is defined. Finally the aircraft noise is detected by comparing threshold with the maximum sum of tendencies that all points direct to the estimated aircraft position. The noise data of the experiment is got by the monitors lay around a large domestic airport and the experiment shows that the algorithm can detect aircraft even it is 1000 m away from the monitoring area and the trace of the aircraft can be reappeared roughly.

Long Short-Term Memory network have been successfully applied to sequence modeling task and obtained great achievements. However, Chinese text contains richer syntactic and semantic information and has strong intrinsic dependency between words and phrases. In this paper, we propose Bidirectional Long Short-Term Memory (BLSTM) with word embedding for Chinese sentiment analysis. BLSTM can learn past and future information and capture stronger dependency relationship. Word embedding mainly extract words’ feature from raw characters input and carry important syntactic and semantic information. Experimental results show that our model achieves 91.46 % accuracy for sentiment analysis task.

One of the dominant problems facing Named Entity Recognition is that when a system trained on one domain is applied to a different domain, a substantial drop in performance is frequently observed. In this paper, we apply active learning strategies to domain adaptation for named entity recognition systems and show that adaptive learning combining the source and target domains is more effective than non-adaptive learning directly from the target domain. Active learning aims to minimize labeling effort by selecting the most informative instances to label. We investigate several sample selection techniques such as Maximum Entropy and Smallest Margin and apply them to the ACE corpus. Our results show that the labeling cost can be reduced by over 92 % without degrading the performance.

Tweet sentiment analysis has been an effective and valuable technique in the sentiment analysis domain. We conduct a systematic and thorough empirical study on traditional machine learning algorithms and two deep learning approaches for tweet sentiment analysis, and expect to provide a guideline for choosing which efficient classification algorithms. Based on our experiments, we found that the Support Vector Machine and the Random Forest work better statistically than other methods. Although deep learning approaches have achieved many successes in image and voice processing, simple RNN and LSTM networks do not outweigh SVM and RF in our experiments. Moreover, for the tweet feature selection, the combination of bi-grams, SentiWordNet and Stop words removal shows more effectiveness in accuracy improving.