Advances in Internetworking, Data & Web Technologies

In order to enhance the detection rate of ensemble classifiers in steganalysis, concern the problems that the accuracy of basic classifier is low and the kind of basic classifier is single in typical ensemble classifiers, an algorithm based on rotating forest transformation and multiple classifiers ensemble is proposed. First, some feature subsets generated randomly merger with training sample to generate sample subsets, then the sample subset is transformed by rotating forest algorithm and train some basic classifiers, which is made of fisher linear discriminate, extreme learning machine and support vector machine with weighted voting. At last, the majority voting method is used to integrate the decisions of base classifiers. Experimental results show that by different steganography approaches and in different embedding rate conditions, the error rate of proposed method decreased by 3.2% and 1.1% in compared with the typical ensemble classifiers and ensemble classifiers of extreme learning machines, therefore demonstrating the proposed method could improve the detection accuracy of ensemble classifier.

Reversible data hiding in encrypted domain (RDH-ED) is an important and effective technical approach for security data management of cloud computing, big data and privacy protection. This paper proposes a three-dimensional (3D) optical reversible data hiding (RDH) with integral imaging cryptosystem. The secret data is encrypted and embedded into the cover image. The receivers can decrypt the cover image and secret data with a reversible or lossless manner, respectively. The simulation experiment and results show that the data embedding rate can be increased to one. Besides, the quality of image decryption is quite high. The technique boasts the advantages of high data embedding rate, security level and real-time capability.

This paper proposes an innovated approach of risk assessment of borrowers based on the BP neutral network model. Specifically, firstly, referring to the empirical data published by the website ‘peer-to-peer lender’ and the indicators of personal credit risk assessment from commercial bank is an efficient method to pick several valid values through data processing, classification and quantification, then the final modeling indicators are selected by information gain technology. Secondly, the new credit risk assessment model is formed after training the modeling indicators. Meanwhile, several strings of collected testing data would be substituted to find out the default rates which are supposed to be compared with the practical ones on the website and the calculated ones from existing credit risk assessment evaluating models. Last but not the least, the effect of this new method is evaluated.

A Wireless Sensor and Actor Network (WSAN) is a group of wireless devices with the ability to sense physical events (sensors) or/and to perform relatively complicated actions (actors), based on the sensed data shared by sensors. In order to provide effective sensing and acting, a coordination mechanism is necessary among sensors and actors. This coordination can be distributed-local coordination among the actors or centralized coordination from a remote management unit, usually known as sink in Wireless Sensor Networks (WSNs). In this work, we propose a simulating system based on Rust for actor node placement problem in WSAN, while considering different aspects of WSANs including coordination, connectivity and coverage. We describe the implementation and show the interface of simulation system. We evaluated the performance of the proposed system by a simulation scenario considering WSANs. The simulation results show that the constructed WSAN could cover both events.

In this research, the primary focus is on privacy preservation in data mining. In particular, the problem of privacy preservation is addressed when the data is to be provided for applications or association rule mining is to be carried out on the datasets shared among two parties, i.e. the two party case. These scenarios are complex to address since privacy issues also lead to the non availability of correct data; also one must meet privacy requirements accompanied by valid data mining results. A system is proposed that is capable of hiding the sensitive information in the given set of data with the help of cryptographic algorithms. The encrypted data is then analyzed using Apriori algorithm for finding frequent itemsets that can lead to vital business decisions. Results reveal that our system provides strong privacy, guarantees accurate data mining while protecting sensitive information during association rule mining.

Cloud computing security is the key bottleneck that restricts its development, and access control on the result of cloud computing is a hot spot of current research. Based on the somewhat homomorphic encryption BGN and combined with Green’s scheme that proposed outsourcing the decryption of CP-ABE (Ciphertext-Policy Attribute-Based Encryption) ciphertexts, we constructed a BGN type outsourcing the decryption of CP-ABE ciphertexts. In our construction, partial decryption of ciphertexts is outsourced to the cloud, and only users whose attribute meets the access policy will get the correct decryption. And the scheme supports arbitrary homomorphic additions and one homomorphic multiplication on ciphertexts. Finally, we prove its semantic security under the subgroup decision assumption and compare it with other schemes.

Wireless Mesh Networks (WMNs) have many advantages such as low cost and increased high speed wireless Internet connectivity, therefore WMNs are becoming an important networking infrastructure. In our previous work, we implemented a Particle Swarm Optimization (PSO) based simulation system for node placement in WMNs, called WMN-PSO. Also, we implemented a simulation system based on Hill Climbing (HC) for solving node placement problem in WMNs, called WMN-HC. In this paper, we implement a hybrid simulation system based on PSO and HC, called WMN-PSOHC. We compare WMN-PSO with WMN-PSOHC by conducting computer simulations. The simulation results show that the WMN-PSOHC has better performance than WMN-PSO.

Mobile computing has many application domains. One important domain is that of mobile applications supporting collaborative work. In a collaborative work, the members of the team has to take decision or solve conflicts in project development (such as delays, changes in project schedule, task assignment, etc.) and therefore members have to vote. Voting can be done in many ways, and in most works in the literature consider majority voting, in which every member of the team accounts on for a vote. In this work, we consider a more realistic case where a vote does not account equal for every member, but accounts on according to member’s active involvement and reliability in the groupwork. We present a voting model, that we call qualified voting, in which every member has a voting score according to four parameters: Number of Activities the Member Participates (NAMP), Number of Activities the Member has Successfully Finished (NAMSF), Number of Online Discussions the Member has Participated (NODMP), Number of Activities the Member Failures (NAMF). Then, we use fuzzy based model to compute a voting score for the member. In this paper, we present two fuzzy-based voting systems (calles FVS1 and FVS2). We make a comparison study between FVS1 and FVS2. The simulation results show that with increasing of the number of activities the member failures, the VS is decreased. When NAMP, NAMSF and NODMP are high, the voting sore is high. The proposed system can choose peers with good voting score in P2P mobile collaborative team. Comparing the complexity of FVS1 and FVS2, the FVS2 is more complex than FVS1. However, it considers also the number of activities the member failures which makes the voting process better.

Due to the amount of anonymity afforded to users of the Tor infrastructure, Tor has become a useful tool for malicious users. With Tor, the users are able to compromise the non-repudiation principle of computer security. Also, the potentially hackers may launch attacks such as DDoS or identity theft behind Tor. For this reason, there are needed new systems and models to detect the intrusion in Tor networks. In this paper, we present the application of Autoregression Integrated Moving Average (ARIMA) for prediction of user behavior in Tor networks. We constructed a Tor server and a Deep Web browser (Tor client) in our laboratory. Then, the client sends the data browsing to the Tor server using the Tor network. We used Wireshark Network Analyzer to get the data and then used the ARIMA model to make the prediction. The simulation results show that proposed system has a good prediction of user behavior in Tor networks.

Ambient intelligence (AmI) deals with a new world of ubiquitous computing devices, where physical environments interact intelligently and unobtrusively with people. AmI environments can be diverse, such as homes, offices, meeting rooms, schools, hospitals, control centers, vehicles, tourist attractions, stores, sports facilities, and music devices. In this paper, we present the implementation and evaluation of actor node an AmI testbed using Raspberry Pi mounted on Raspbian OS. For evaluation, we considered respiratory rate and heart rate metrics. We carried out an experiments and clustered sensed data by k-means clustering algorithm. From experimental results, we found that the implemented AmI testbed gives a good effect to human during sleeping.

Encryption algorithms in Internet of Things are a piece of small area with small-scale, it need some light weight encryption algorithms. This paper focuses on the component of encryption algorithms, some light weight of the rotation boolean permutations are perfectly characterized by the matrix of linear expressions. Three methods of rotation nonlinear boolean permutations are constructed. The sub-functions of the three permutations have three monomials, hight degree, 2-algebra immunity. All three classes of rotation nonlinear boolean permutations are fully determination by the first component Boolean function, respectively.

Patent analysis and mining can excavate valuable information hidden in patent texts, and help enterprises to make correct decisions. As an important step in patent mining, whether patent semantic annotation is correct or not directly affects the results of mining. During the annotation of effect statements, whether manual or automatic, we need to use clue words to judge. This paper presents a construction method of clue words thesaurus in Chinese patents based on iteration and self-filtering, in order to improve the accuracy of effect statements’ annotation.

With the rapid development of computing power, numerical simulation has become useful and powerful tool. Expanded mixed finite element method is introduced to solve the nonlinear elliptic problem in divergence form. Existence and uniqueness of the discrete problem are demonstrated. Optimal L$$^2$$-error estimates for three variables are got. Numerical examples are provided to validate the theoretical analysis.

This paper proposes an improved version of Shiu’s encrypted image-based reversible data hiding with public key cryptography (EIRDH-P). The original work vacates embedding room by difference expansion technique and embeds one bit into each pair of adjacent encrypted pixels. The data extraction and image recovery can be achieved by comparing all pairs of decrypted pixels. Shius’ work did not fully exploit the correlation inherent in the neighborhood of a pixel and required side information to record the location map. These two issues could reduce the amount of differences and in turn lessen the potential embedding capacity. This letter adopts a better scheme for vacating room before public key encryption using prediction-error expansion method, in which the pixel predictor is utilized by interpolation technique. The experimental results reveal that the proposed method offers better performance over Shiu’s work and existing EIRDH-P schemes. For example, when the peak signal-to-noise ratio of the decrypted Lena image method is 35, the payload of proposed method is 0.74 bpp, which is significantly higher than 0.5 bpp of Shius’s work.

This paper proposes a novel reversible data hiding algorithm with image contrast enhancement based on homomorphic public key cryptosystem. The additional data is embedded based on histogram shifting after preprocessing procedure. Then the image is encrypted using public key and side information is embedded. On the receiver side, the image with contrast enhancement is obtained directly after image decryption using private key. Due to the correlation between adjacent pixels, date extraction and image recovery can be implemented. To our best knowledge, it is the first reversible data hiding in encrypted image algorithm with image contrast enhancement. Experimental results have demonstrated the feasibility and effectiveness of the proposed method.

This paper presents a new deep network (non – very deep network) with composite residual for handwritten character recognition. The main network design is as follows: (1) Introduces an unsupervised FCM clustering algorithm to preprocess the experimental data. (2) By exploiting a composite residual structure the multilevel shortcut connection is proposed which is more suitable for the learning of residual. (3) In order to solve the problem of overfitting and time-consuming for training the network parameters, a dropout layer is added after the completion of all convolution operations of each extended nonlinear residual kernel. Comparing with general deep network structures of same deep on handwritten character MNIST database, the proposed algorithm shows better recognition accuracy and higher recognition efficiency.

Binary hashing has been widely used for efficient similarity search due to its query and storage efficiency. In this paper, we attempt to exploit ensemble approaches to tackle hashing problem. A flexible ensemble hashing framework is proposed to guide the design of hashing methods, which takes into account three important principles namely higher accuracy, larger diversity and the optimal weights for predictors simultaneously. Next, a novel hashing method is designed by the proposed framework. In this work, we first use the weighted matrix to balance the variance of hash bits and then exploit bagging method to inject the diversity among hash tables. Under the same code length, the experimental results show that the proposed method achieves better performance than several other state-of-the-art methods on two image benchmarks CIFAR-10 and LabelMe.

With the rapid development of computer technology, educational information technology plays a more and more important role in modern education. The development of cloud computing technology has brought great influence and change to the construction of network learning platform. Aiming at the problem of design and implementation under the environment of network learning platform for cloud computing, through in-depth analysis of the current network learning under cloud computing environment, learning theory, learning object and system function, system construction is given a practical significance of the cloud computing network learning platform and system design method, provides the theoretical basis and reference for promoting the application of cloud computing technology in modern education.

Due to the large volume of text documents uploaded on the Internet daily. The quantity of Kurdish documents which can be obtained via the web increases drastically with each passing day. Considering news appearances, specifically, documents identified with categories, for example, health, politics, and sport appear to be in the wrong category or archives might be positioned in a nonspecific category called others. This paper is concerned with text classification of Kurdish text documents to placing articles or an email into its right class per their contents. Even though there are considerable numbers of studies directed on text classification in other languages, and the quantity of studies conducted in Kurdish is extremely restricted because of the absence of openness, and convenience of datasets. In this paper, a new dataset named KDC-4007 that can be widely used in the studies of text classification about Kurdish news and articles is created. KDC-4007 dataset its file formats are compatible with well-known text mining tools. Comparisons of three best-known algorithms (such as Support Vector Machine (SVM), Naïve Bays (NB) and Decision Tree (DT) classifiers) for text classification and TF × IDF feature weighting method are evaluated on KDC-4007. The paper also studies the effects of utilizing Kurdish stemmer on the effectiveness of these classifiers. The experimental results indicate that the good accuracy value 91.03% is provided by the SVM classifier, especially when the stemming and TF × IDF feature weighting are involved in the preprocessing phase. KDC-4007 datasets are available publicly and the outcome of this study can be further used in future as a baseline for evaluations with other classifiers by other researchers.

With the proliferation of computation and storage outsourcing, access control has become one of the vital problems in cloud computing. Supporting more complex and flexible function, predicate encryption (PE) is gaining more and more attention in access control of decryption outsourcing. Based on the somewhat homomorphic encryption Paillier scheme and combined with Green’s scheme which supports outsourcing the decryption ciphertexts, we constructed a Paillier type outsourcing the decryption ciphertexts via predicate encryption. In this scheme, decryption can be outsourced to the cloud, and which can greatly reduce the storage and computation costs of user end. Moreover, the scheme supports arbitrary homomorphic addition and one homomorphic multiplication on ciphertexts. IND-AH-CPA security of the scheme is proved under subgroup decision assumption.

With the developments of RFID Technology, the unreliability of RFID devices result in a large number of tag data’s redundancy data, and the reliability of RFID source data is gradually harder to get the safeguard. The reliability means the correctness of the RFID source data such as multi and leakage of reading data. This paper designs a new RFID middleware architecture, which introduces the system structure of middleware and shows the details of each significant part of system. The data processing module and data transmission module are the two main parts designated to get reliable RFID data. The data processing module uses L-NCD algorithm to process data redundancy. The data transmission module takes the responsibility of transferring data between RFID middleware and the sever through JMS message transferring mechanism.

The PRINCE is a light-weight block cipher with the 64-bit block size and 128-bit key size. It is characterized by low power-consumption and low latency. PRINCEcore is the PRINCE cipher without key-whiting. For evaluating its security, a statistical testing on linear transformation is performed, and a statistical character matrix is given. By using the “miss-in-the-middle” technique, we construct 5-round zero-correlation linear approximations. Based on the 5-round distinguisher, a 9-round attack on the PRINCEcore is performed. The data complexity is $$ 2^{62.9} $$ known plaintexts and the time complexity is $$ 2^{55.14} $$ 9-round encryptions. The testing result shows that the PRINCEcore reduced to 9 rounds is not immune to multidimensional zero-correlation linear analysis.

Distance Measuring Equipment/Precision (DME/P) provides precisely distance for the approaching plane. The Simulated DME/P Beaconing System is designed for testing the performances index of airborne facility and researching interference of white noise and pulse in laboratory environment. Particular design scheme is raised,and implementation based on FPGA is discussed in this paper. The simulation results indicate that this system can satisfy for testing requirement and also quantify the influence of interference.

It is necessary to use the channel parameters to initialize the iterative decoding in sum-product decoding algorithm of LDPC codes. The channel parameters of the receiver are unknown, so it is essential to estimate the channel parameters or the signal-to-noise ratio (SNR). In this paper, two estimation models are established by the characteristics of the Gaussian channel, and their precision are analyzed; then we use the moment estimation, maximum likelihood estimation and Bayesian estimation respectively to estimate the models mentioned above; finally use the estimated signal-to-noise ratio to decode the LDPC codes by sum-product decoding. Simulation results show that: the result of the estimation model 2 is more accurate than that of the estimation model 1. Using the result of the estimation model 2 to decode the LDPC codes by sum-product decoding, the curve of the decoding results and that of using true value to decode are almost overlapped.

Information retrieval (IR) is a major technology helping people to retrieve the information they are interesting in. One of challenge in IR is that the input query consists of very few words so that IR can’t catch the user’s intention. The pseudo correlation query extension (PCQE) is a power technology in IR aim to solve this challenge. In this paper, we propose a PCQE method which based on LDA, it apply the LDA model to fit the document set, then the latent topics are exploited and each document is represented as a multinomial distribution over topics. We calculate the probability of the document generating the query to measure the correlation between them, then the documents are ranked in terms of the correlation and top documents are extracted to seek informative words to extend the original query. Our experiment on the Ohsumed data set shows our method outperforms the other state-of-art PCQE methods.

In order to conquer the hard outputs defect of Support Vector Machine (SVM) and extend its application, an improved target recognition method based on Multi-class Support Vector Machine (MSVM) is proposed. Firstly, the typical Probability Modeling methodologies of MSVM were deeply analyzed. Secondly, the structure of one-against-one multi-class method which matches with Basic Probability Assignment (BPA) outputs of evidence theory by coincide, so a special Multi-class BPA output method is derived, and multi-sensor target recognition model based on MSVM and two-layer evidence theory is constructed. Finally, the results of experiments show that the proposed approach can not only conquer the overlap area of one-against-one multi-class method, but also improve classification accuracy.

In order to overcome the disadvantages of high complexity, low speed and accuracy of selective ensemble algorithm based on genetic algorithm. We proposed a new selective ensemble algorithm based on probability PSO algorithm. First, in order to tackle the converging slowly and easily to partial minimum problems of simplified PSO, we introduce the cloud model and the method of complex; Second, we present the definition of probability PSO and the formula that converting the particle position vector to base learner selection problem, that make the transformation from continuous space to discrete space become true. Finally, we choose integration model generalization error as the adaptive function of PPSOSEN. The numerical results show that, compared with discrete PSO, PPSOSEN improved the recognition precision with the same time consumption, and it is an efficient selective ensemble algorithm.

With the rapid development of cloud computing, more and more consumers adopt third-party cloud services to implement their critical business. Cloud services are provided by different service providers, and usually have different cost performance and quality of service. Service consumers have to make trade-offs on multiple factors in order to choose the appropriate cloud services. Therefore, an efficient cloud service selection mechanism makes sense for both cloud service providers and consumers. According to existing works, this research field still requires appropriate models, effective design paradigm, in-depth experimentations and practical implementations. Because agent can identify the paradigm of the clouds by learning algorithm, they can be trained to observe the differences and behave flexibly for cloud service selection. To rank different clouds, we propose and assign QoS factor for each cloud service and ranks it as whole. According to simulation experiment, we validate the approach, which emphasizes the need to rank cloud services of widely spreading and complex domains.

Aimed at developing and utilizing the water resource appropriately, it is critical to analyze, mine and present the valuable information and knowledge. Until recently, analyzing the big data in an online environment has not been an easy task especially in the eyes of data consumers in water conservancy domain. Moreover, there is no single tool or a one-size-fits-all solution for big data processing and data visualization in a special field. This barrier is now overcome by the availability of cloud computing, R and Shiny. In this paper, we propose to develop cloud-based tools for water resource data analysis using R and Shiny. Following the whole solution, the implementation using long-term hydrological data collected from Chu River is introduced as an example. The results show that these tools are valuable and practical resource for individuals with limited web development skills and offer opportunity for more dynamic and collaborative water resource management.

Unknown network protocol’s dormant behavior is becoming a new type of stealth attack, which greatly harms the cyber space security, and seriously affects the credibility of network protocols. By studying the characteristics of the dormant behavior, we discovered that protocol’s dormant behavior is different from the normal one in instruction level. The protocol’s behaviors can be represented by the labeled instruction sequences. The similar instruction sequences means the similar protocol behavior, but the dormant behaviors are the ones that can only be triggered under particular conditions. The main frame of perception mining network protocol’s dormant behavior, and sensitive information automatic generation are proposed in this paper. Using our proposed instruction clustering algorithm, all kinds of executed or unexecuted instruction sequences can automatic clustering. On the basis of this, we can effectively distinguish and mine the protocols’ potential dormant behaviors. A virtual protocol behavior analysis platform HiddenDisc has been developed, and the protocol execution security evaluation scheme is proposed and implemented. Through the analysis and evaluation of the 1297 protocol samples, the experimental results show that the present method can accurately and efficiently perception mining unknown protocol’s dormant behaviors, and can evaluate the protocol’s execution security.

The camera systems are usually suffered from random jitter. In this paper, a new method based on Kalman filter and homography transformation is proposed to stabilize the unstable video. Firstly, the SURF (Speed-Up Robust Feature) point-feature matching algorithm is employed to find the corresponding matching points between two consecutive frames, and the bidirectional nearest neighbor distance ratio method is used to clear false matches. Secondly, motion estimation is computed by homography model and least square method. Then, Kalman filter are applied to separate the global and local motion. Finally, the unstable video frames is compensated by global motion vector. The experiment result shows that proposed method can effectively eliminate the random jitter.

This paper presents the first results of a teaching innovation project named “Emotional Thermometer for Teaching” (ETT) carried out at the Universitat Oberta de Catalunya. The ETT project intersects the scopes of eLearning and Affective Computing with the aim of collecting and managing emotional information of online students during their learning process. Such information allows lecturers to monitor the overall emotional climate of their virtual classrooms whilst detecting critical moments for timely interventions, such as assisting in certain learning tasks that generate negative emotions (anxiety, fear, etc.). To this end, an innovative teaching tool named ETT was developed as a functional indicator to measure and represent the classroom emotional climate, which is dynamically evolving as the course goes by. In this paper, the technical development of the ETT tool is described that meets the challenging requirement of correctly identifying the overall emotional climate of virtual classrooms from the posts sent by students to in-class forums. First, a machine learning approach combined with Natural Language Processing techniques is described to automatically classify posts in terms of positive, neutral and negative emotions. Then, a web-based graphical tool is presented to visualize the calculated emotional climate of the classroom and its evolution over time. Finally, the post classification approach is technically tested and the initial results are discussed.

Bilinear pairing computation is one of the most important cryptographic primitives, which is widely used in the public key encryption schemes. However, it has been considered the most expensive operation in the pairing-based cryptographic protocols. In this paper, we present an efficient and secure outsourcing algorithm for bilinear maps based on one untrusted servers. The client could outsource expensive computation to the cloud and perform simple operation to obtain the great efficiency. We analyze the security of this algorithm and compare it with prior works in efficiency. It is argued that our algorithm is more efficient and practical than the state of the art.

Broadcasters use broadcast encryption to broadcast confidential communications to arbitrary sets of users, and broadcasters individually send their corresponding ciphertext information for different sets of users. However, in the modern Internet, which is represented by cloud computing and complex networks, with the rapid increase of broadcast users and the rapid growth of the amount of broadcast information, the number of broadcast users is increasing. In order to solve this problem, a broadcast encryption scheme is proposed. In the environment of multi user collection, the new scheme has good communication and computation overhead, and the ciphertext length is only constant. The new scheme is flexible and efficient, and can be widely used in many fields, such as pay TV.

The key encapsulation mechanism (KEM) and the data encapsulation mechanism (DEM) form a hybrid encryption, which effectively solves the problem of low efficiency of public key cryptography and key distribution problems in symmetric encryption system. The security and efficiency of the key encapsulation mechanism directly affect the security and efficiency of hybrid encryption. In this paper, an identity-based key encapsulation scheme is constructed by using multilinear mapping. We proved that the scheme is under the standard model of adaptive chosen-ciphertext security. The scheme can be publicly verified and the key and ciphertext length are constant and have high efficiency.

Many applications in VANETs rely on reliable broadcast, while broadcasting can led to broadcast storm problem. Probabilistic broadcast is a kind of simple and effective way to suppress broadcast storm. However, a lot of existing probabilistic broadcast protocols for VANETs haven’t taken network division problem in low traffic density into consideration. This paper design and implement an enhanced local connectivity-based broadcast protocol. Simulation results show that the new protocol can achieve better reliability with lower overhead in both dense and sparse traffic scenario.

k-means is one of the most widely used clustering algorithms by far. However, when faced with massive data clustering tasks, traditional data mining approaches, especially existing clustering mechanisms fail to deal with malicious attacks under arbitrary background knowledge. This could result in violation of individuals’ privacy, as well as leaks through system resources and clustering outputs while untrusted codes are directly performed on the original data. To address this issue, this paper proposes a novel, effective hybrid k-means clustering preserving differential privacy in Spark, namely Differential Privacy Hybrid k-means (DPHKMS). We combined Particle Swarm Optimization and Cuckoo-search to initiate better cluster centroid selections in the framework of big data computing platform, Apache Spark. Furthermore, DPHKMS is implemented and theoretically proved to meet ε-differential privacy with determinative privacy budget allocation under Laplace mechanism. Finally, experimental results on challenging benchmark data sets demonstrated that DPHKMS, guaranteeing availability and scalability, significantly improves existing varieties of k-means and consistently outperforms the state-of-the-art ones in terms of privacy-preserving, verifying the effectiveness and advantages of incorporating heuristic swarm intelligence.

Image fusion has been a hotspot in the area of image processing. How to extract and fuse the main and detailed information as accurately as possible from the source images into the single one is the key to resolving the above problem. Convolutional neural network (CNN) has been proved to be an effective tool to cope with many issues of image processing, such as image classification. In this paper, a novel image fusion method based on pulse-coupled neural network (PCNN) and CNN is proposed. CNN is used to obtain a series of convolution and linear layers which represent the high-frequency and low-frequency information, respectively. The traditional PCNN is improved to be responsible for selecting the coefficients of the sub-images. Experimental results indicate that the proposed method has obvious superiorities over the current main-streamed ones in terms of fusion performance and computational complexity.

Based on the traditional iterative reconstruction workflow, a fast iterative reconstruction algorithm FIRA is proposed. First, using the image feature points extracted by SIFT algorithm, calculation of image similarity based on the minimum hash algorithm in LSH model is performed. Then, the iteration order is specified through hierarchical clustering. In the iterative process, the orientation estimation of images is carried through the clustering result coming from hierarchical tree. The optimization of parameter estimation is performed by bundle adjustment, and finally produce 3d mesh models. The experimental results show that the method could bring high efficiency and eliminate the accumulated error of adjustment calculation.

Replacement algorithm optimization is the core of cache model research. On the basis of the cache replacement model RFS, through long-term observation and analysis to the real network logs, find that the fluctuation of the access interval change rate is more valuable in predicting the new objects arrival. Therefore, in this paper, we first get the access heat level through clustering the access interval change rate with the improved K-means clustering algorithm; and then establish HSF optimal web cache model with the access heat level which named H, web object size which named S and web object freshness which named F. The replacement strategy of HSF model’s is: First, replace the lowest heat level of the web object; replace the biggest size one, if H is the same; replace The lowest freshness one if H and S are the same. The simulation shows that the HSF model had the better hit rate and the byte hit rate, and the lower the access delay than the RFS.

The growth of online crowdsourcing marketplaces has attracted massive normal buyers and micro workers, even campaigners and malicious users who post spamming jobs. Due to the significant role in information seeking and providing, CQA (Community Question Answering) has become a target of crowdsourcing spammers. In this paper, we aim to develop a solution to detect crowdsourcing spammers in CQA websites. Based on the ground-truth data, we conduct a hybrid analysis including both non-semantic and semantic analysis with a set of unique features (e.g., profile features, social network features, content features and linguistic features). With the help of proposed features, we develop a supervised machine learning solution for detecting crowdsourcing spammers in Community QA. Our method achieves a high performance with an AUC (area under the receiver-operating characteristic curve) value of 0.995 and an $$F_{1}$$ score of 0.967, which significantly outperforms existing works.

In recent years, we have witnessed a dramatic growth in spam mail. Other related forms of spam are also increasingly exposed the seriousness of the problem, especially in the short message service (SMS). Just like spam mail, the problem of spam message can be solved with legal, economic or technical means. Among the technical means, Bayesian classification algorithm, which is simple to design and has the higher accuracy, becomes the most effective filtration methods. In addition, from the perspective of social development, digital evidence will play an important role in legal practice in the future. Therefore, spam message, a kind of digital evidence, will also become the main relevant evidence to the case. This paper presents a spam message detection model based on the Bayesian classification algorithm. And it will be applied to the process of SMS forensics as a means to analyze and identify the digital evidence. Test results show that the system can effectively detect spam messages, so it will play a great role in judging criminal suspects, and it can be used as a workable scheme in SMS forensics.

In recent years, e-mail technology is prospering, bringing efficiency to people from all over the world. It is not limited to time and space, making the transmission of information more convenient. However, the emergence of spam has also brought people a lot of trouble. Thus, spam filtering research is necessary. Traditional spam filtering is mainly based on black and white list technology. Over the past decade, with the development of machine learning, Bayesian classifier has also come into use. However, support for Chinese mail has always been unsatisfactory. This paper proposes a Chinese spam filtering method based on suffix tree, which solves the problem of Chinese character processing and compares it with traditional methods from the aspects of time and space complexity and accuracy.

This paper analyzes the current situation of MP3 application and the advantages of wavelet transform in audio watermarking, and effectively uses of the unipolar quantization method to propose an MP3 audio watermarking algorithm based on wavelet transform. The algorithm firstly decodes the MP3 audio, then uses the third-order discrete wavelet transform, embeds the low-frequency coefficients with the method of unipolar quantization, and finally obtains the watermarked MP3 audio file. Simulation results show that the algorithm has good auditory transparency and strong robustness by low-pass filtering, resampling, whitening, and cropping attacks on watermarked audio signals. The watermarking can be fast extracted, and meet the real-time requirements.

In the group argumentation environment, a large amount of text information will be produced. How to find the specific speeches of experts from many similar speeches and extract their common summary is of great significance to improve the efficiency of experts’ argumentation and promote consensus. In this paper, the heuristic method is first used to cluster the speech texts and find the similar speech sets. Then, we use TextRank algorithm to extract multiple document summary, and feedback the summary to the experts. The experimental results show that the efficiency of the experts’ argumentation is improved and the decision-making is promoted.

This paper a novel face detection method based on visual saliency mechanism to improve the accuracy of unconstrained face recognition. Log Gabor transformation is used to extract visual features, and obtain facial saliency map by using stable balance measurement method based on Graph-Based Visual Saliency. Then binary image is obtained by segmenting facial saliency map with maximum entropy threshold and the rectangle area is marked by setting the centroid of object region as the center. Face region is detected from the original image according to the rectangle area. Experimental results on LFW database show that our algorithm can effectively remove the background interference without losing any face information and quickly precisely detect the face region which is more conducive to the unconstrained face recognition.

Pavement crack detection plays an important role in pavement maintaining and management. In recent years, pavement crack detection technique based on range image is a recent trend due to its ability of discriminating oil spills and shadows. Existing pavement crack detection methods cannot effectively detect transverse and network cracks, because these methods generally represent the crack geometry feature using single laser scan line, which cannot take the effects of spatial variability, anisotropy and integrity into account. Aiming at the deficiency of existing algorithms, the pavement crack detection method fused histogram of oriented gradient and watershed algorithm is proposed. Firstly, crack edge strength and orientation are detected by histogram of oriented gradient in pavement range image. Then, the traditional watershed algorithm is improved by using the crack edge orientation in order to better extract the crack object. Experiment results show that the proposed method can accurately detect different types of crack objects and identify the severity of crack damage simultaneously.

This paper proposes a novel framework of multi-hypothesis compressed video sensing. Multi-hypothesis prediction and bi-directional motion estimation are applied to generate side information candidates. The correlation coefficients between the non-key frame and the three candidates are calculated respectively for selecting the most similar side information. The simulation results show that the proposed framework can provide better recovery performance than the framework using original MH-BCS-SPL algorithm.

Three-party password-based authenticated key exchange (3PAKE) protocol allows two clients, each sharing a password with a trusted server, to establish a secret session key with the help of the server. It is a practical mechanism for establishing secure channels in the communication networks. Recently, Xu et al. proposed a 3PAKE protocol without the server’s public key. They claimed that their protocol could withstand various attacks. In this paper, we show Xu et al.’s protocol is insecure against the stolen-verifier attack. Furthermore, we propose an improved 3PAKE protocol to overcome the weakness of Xu et al.’s protocol. Security and performance analysis shows that our protocol not only overcomes the security weakness, but also is more efficient. Therefore, our protocol is more suitable for the practical applications.

Network coding is theoretically the most efficient coding scheme for decentralized networks with better throughput and better robustness. However, if malicious intermediate nodes launch pollution attacks to the data by triumphantly forging network code, the sink node would suffer from failed decoding with bandwidth wasting, longer delay and more overheads. The classic bit by bit digital signature schemes are elegant, but the computation complexity is high, for each bit have to execute a hash computation. The pollution detection schemes based on null key cannot against colluding attacks. The schemes based on homomorphic MAC ensure the sink nodes verify the data, but those intermediate nodes cannot detect the pollution attacks. Above schemes are not enough efficient. In this paper, we propose a new combined security network coding scheme based on homomorphic MAC and null key that overcome the shortage of each other.

This passage puts forward a new sparse representation method, to solve the shortage problem of image restoration. First of all, extract the patch groups by utilize the non-local similar patches, and then using the simultaneous sparse coding to develop a non-local extension of Gaussian scale mixture model. Finally integrate the patch group model and Gaussian scale mixture model into encoding framework. Experimental results show that the proposed method achieves leading performance in terms of both quantitative measures and visual quality. In addition, our algorithm generates high-resolution images that are competitive or even superior in quality to images produced by other similar methods.

Network coding is now widely used to improve the network throughput capacity in lots of applications, such as distributed storage, wireless mesh networks, etc. Unlike the traditional routing scheme in which the network nodes simply relay the received packets, network coding technique requires the intermediate node to combine the received packets together and then re-transmit it repeatedly. However, there is a fatal threat that the malicious intermediate nodes can tamper the data before combining the packets, and thus the standard signature scheme cannot satisfy the security requirement for this application. In this paper, we propose an identity-based homomorphic scheme for network coding which can prevent malicious nodes to produce the pollution attacks. The public key of our scheme is a constant size which is only the hash output of user’s identity. We present the detailed construction and analyze the security of the scheme in the random oracle model.

This paper presents a three-dimensional (3D) digital watermarking technique based on integral image cryptosystem and discrete Fresnel diffraction (DFD). 3D digital watermarking is generated by computational integral imaging. The 3D digital watermarking is encrypted and embedded by integral imaging cryptosystem that is designed with DFD transform algorithm. Finally, the extracted 3D digital watermarking is decrypted and displayed by integral imaging cryptosystem. The feasibility and effectiveness of the proposed scheme is demonstrated by numerical simulation experiment. The majority of system will improve the security and robustness of 3D digital watermarking. The proposed method can provide a new, real-time, and effective strategy in the security data management of cloud computing and big data.

Existing travel related systems and commonly used websites have some major limitations which cause efforts to be made by the traveler before going out on vacation. Some of these sites allow users to write their personal experiences about visited places but don’t produce a proper itinerary, and those which do, focus only on minimizing the travel time between POIs ignoring other important factors like POI ratings, traffic conditions, etc.Our work focuses on Building Real-Time Travel Itineraries using ‘off-the-shelf’ data from the Web. The proposed solution solves the existing limitations by using an optimization algorithm, which produces a real-time itinerary after optimizing various important factors like travel time between POIs, traffic conditions, ratings of POIs, to enhance the traveler’s experience in a city.Out of the several optimization approaches available, an algorithm was finalized after comparison of performance and accuracy between the approaches. Best results were obtained in case of a dynamic programming based approach, which optimized both accuracy and performance.

With the emergence of smart grid (SG), the residents have the opportunity to integrate renewable energy sources (RESs) and take part in demand side management (DSM). In this regard, we design energy management control unit (EMCU) based on genetic algorithm (GA), binary particle swarm optimization (BPSO), and wind driven optimization (WDO) to schedule appliances in presence of objective function, constraints, control parameters, and comparatively evaluate the performance. For energy pricing, real time pricing (RTP) plus inclined block rate (IBR) is used. RESs integration to SG is a challenge due stochastic nature of RE. In this paper, two techniques are addressed to handle the stochastic nature of RE. First one is energy storage system (ESS) which smooths out variation in RE generation. Second one is the trading/cooperation of excess generation to neighboring consumers. The simulation results show that WDO perform more efficiently than unscheduled in terms of reduction in: electricity cost, the tradeoff between electricity cost and waiting time, and peak to average ratio (PAR). Moreover, incorporation of RESs into SG design increase the revenue and reduce carbon emission.

In smart grid, several optimization techniques are developed for residential load scheduling purpose. Preliminary all the conventional techniques aimed at minimizing the electricity consumption cost. This paper mainly focuses on minimization of electricity cost and maximization of user comfort along with the reduction of peak power consumption. We develop a multi-residential load scheduling algorithm based on two heuristic optimization techniques: genetic algorithm and binary particle swarm optimization. The day-ahead pricing mechanism is used for this scheduling problem. The simulation results validate that the proposed model has achieved substantial savings in electricity bills with maximum user comfort. Moreover, results also show the reduction in peak power consumption. We analyzed that user comfort has significant effect on electricity consumption cost.

The performance and comparative analysis of home energy management controller using three optimization techniques; genetic algorithm (GA), enhanced differential evolution (EDE) and optimal stopping rule (OSR) has been evaluated in this paper. In this regard, a generic system model consisting of home area network, advanced metering infrastructure, home energy management controller, and smart appliances has been proposed. Price threshold policy and priority of appliance have also been considered to depict monthly and yearly average electricity bill savings and appliance delay using day-ahead real-time pricing (DA-RTP). Simulation results validate that all our proposed schemes successfully shifts the appliance operations to off-peak times and results in reduced electricity bill with reasonable waiting time.

In order to improve the accuracy of the detection and tracking task in the intelligent surveillance system, we propose a shadow elimination algorithm based on HSV spatial feature and texture feature. In this paper, firstly the background subtraction is used to obtain the motion area of the sequence image, where HSV feature is used to determine the threshold value of the shadow elimination which can be completely removed. Then the complete moving target is obtained by OR operator of combining the foreground which is extracted by OTSU and the result which is extracted by HSV. The algorithm is applied to several realistic scenario where exists various shadow. We compare our method with other traditional algorithm and report experimental results, both in terms of noise suppression and detection accuracy. The experimental results show that the proposed method has the better noise suppression and detection accuracy.

Based on mobile devices limitations, several user authentications and key exchange schemes have been proposed for mobile devices using identity-based public key cryptography (ID-PKC). However, these schemes suffer from key escrow problem. Moreover, they are not secure against impersonation attacks, and they can’t achieve perfect forward secrecy. In this paper, a new user authentication and key exchange protocol for the mobile client-server environment is proposed. Certificateless public key cryptography (CL-PKC) and bilinear pairing are adopted in the proposed scheme. Our protocol solves the key escrow problem of identity-based public key cryptography. Also, it is secure against both adversaries type I and type II. Furthermore, the proposed protocol achieves perfect forward secrecy. We prove the security of our protocol in the random oracle model under the Computational Diffie-Hellman (CDH) problem. Hence, the proposed scheme is more suitable for the mobile devices environments.

Attribute based encryption is a very useful primitive for scalable access control on the ciphertexts and has found broad applications, such as secure cloud storage etc. When this primitive is used by mobile phones, the computation cost is too heavy. So Hohenberger and Waters introduced the concept of Online/offline attribute based encryption and give a such concrete construction. In this paper, we give an improved construction based on their proposal. Compared with their proposal, our proposal needs 5 pairings instead of $$2|I|+1$$ pairings, which is much more efficient than the original scheme. Furthermore, we generalize this technique to speed up the computation of multi-modular exponentiation, and thus also get an interesting result.

Nowadays cloud data storage is a very important storage service for us, but to ensure the datum stored in the remote cloud server remains unmodified, we need a mechanism to check the datum’s integrity, cloud data storage auditing protocol is such a mechanism, which has received great attention from researchers. Recently Zhang et al. proposed an efficient ID-based public auditing protocol called IPAD for the outsourced data by combing Waters signature and public auditing for the outsourced data. They claimed IPAD is the first ID-based auditing protocol for data integrity in the standard security model. But in this paper we show their proposal is not secure. Especially, the adversaries can easily generate tags for any file, which obviously break the unforgeability property of the cloud storage auditing protocol.

The textual data grows explosively with the advent of the era of big data, a significant portion of textual data is text documents labeled with multi-label such as the papers with keywords. Multi-label classification is a power technology to handle the multi-labeled textual data, but a huge room stays for improving the effect of multi-label classifying for textual data. This paper introduces labeled LDA with function terms (LF-LDA), a topic model that extracts noisy function terms from textual data to improve the performance of multi-label classification. The experimental result on RCV1-v2 textual dataset shows that LF-LDA can outperform the other two state-of-art multi-label classifiers: Tuned SVM and L-LDA on both Macro-F1 and Micro-F1 metrics. The low variance also indicates LF-LDA is a robust classifier.

With the development of the social economy and the improvement of the people’s living standard, more and more categories of infant formulas are presented according to nutritional requirements and regional differences. For a specific family, nowadays it is usually quite difficult to make a quick decision. This manuscript firstly analyzes some infant formulas made in Canada, The Netherlands, Denmark, Ireland and Germany, and then outlines the special nutrients of each given kind of infant formula. Based on these observations, dataset construction and classification are discussed so that relational decisions can be made according to specific needs.

As the integration of informatization and industrialization goes deeper in State Grid Jiangsu Electric Power Company, the lean management of O&M (operation and maintenance) of information system plays a more important role in the company’s production and management. On the ground of a full investigation of the current information system of the company, this paper has improved the model of business impact analysis (BIA) and, based on which, proposed a new method to classify O&M of information system. As proved in our practices in the company, the proposed model and method are efficient in controlling the cost of optimizing the operation of the information system, raising the efficiency of resource utilization as well as in improving the O&M management.

Scientific community is now spending more and more efforts in defining and developing effective methodologies and technologies in order to easy design and development of Cloud solutions. In order to exploit the features of existing Cloud services and Resources Orchestration becomes a hot research topic. In this scenario, Cloud Designers promote reuse but a clear and simple design and verification methodology still misses in literature. In this scenario, a simple (UML-based) modelling profile and a Model-Driven Engineering methodology for Cloud-based Value Added Services are very appealing. In this work we define a modelling profile able to describe Orchestrated Cloud Services and Resources by means of Cloud Design Patterns and we show how Cloud Designer can use it both to ease composition and verification purposes.

This paper Propose the location-based routing algorithm forwarding with alternative copies. This routing algorithm considers both the location information and the velocity vector to establish the utility function, which uses a factor of speed to adapt to the change of the node density. And the forwarding strategy of alternative copies can guarantee the success of the transfer between adjacent nodes. The simulation results demonstrate preliminarily that the DTN routing algorithms designed in the paper performs well in terms to the delivery ratio, the delays and the network overhead.

Considering the lack of theoretical analysis for systems under complicated attacks, a framework was proposed to analyze attack risks based on attack-defense trees. The attack period was divided into attack phase and defense phase and metrics was defined. First, action nodes were constructed by collecting system vulnerabilities and capturing invasive events, and defense strategies were mapped to defense nodes in the tree structure. Besides, formal definitions were given and attack-defense tree with metrics was constructed using ADTool and relevant algorithms. In addition, concepts of ROA (Return on attack) and ROI (Return on Investment) were introduced to analyze system risk as well as to evaluate countermeasures. Finally, a risk analysis framework based on attack-defense trees was established and numerical case was given to demonstrate the proposed approach. The result showed that the framework could clearly describe the practical scenario of the interaction between attacks and defenses. The objective of risk analysis and countermeasures evaluation could be achieved.

Multi-focus image fusion is a classic issue in the field of image processing. How to extract and fuse the in-focus information from the source images into the single one is the key to resolving the above problem. As a novel multi-resolution analysis tool, non-subsampled shearlet transform (NSST) not only has better information capturing ability, but also owns a comparatively lower computational complexity compared with non-subsampled contourlet transform (NSCT). Intersecting cortical model (ICM) is the third generation of artificial neural network, and it can be viewed as the improved version of pulse-coupled neural network. The superiority of ICM lies in that it has much fewer parameters and better function mechanism. In this paper, a novel method for multi-focus image fusion based on NSST and improved ICM is presented. On the one hand, NSST is responsible for decomposing source images and reconstructing sub-images. On the other hand, ICM is used to complete the coefficients selecting of sub-images. Experimental results demonstrate that the proposed method has better performance compared with the current typical ones.

The pilot contamination problem has been the primary limitation of massive multiple input multiple output (MIMO) systems. To improve it, in this paper, we propose a dynamical pilot assignment algorithm based on the priority of user location. First, we obtain the formulation of signal to interference plus noise power ratio (SINR) in uplink channel through minimum mean square error (MMSE) mechanism. Second, an objective function of SINR is defined together with constraint condition of real distance, based on which optimal value (OV) could be achieved. Third, we propose a novel cellular classification algorithm, that is, area with better channels adopts random pilot assignment scheme, and others use the novel algorithm. Last, the proposed algorithm is compared with the traditional algorithm. The results show that the proposed algorithm can effectively reduce the influence of pilot contamination on the communication performance and improve the system SINR and the system capacity.

A method based on the leader-follower method is proposed for formation transformation in large-scale mass incidents. The greedy algorithm is introduced to realize regional division and leader matching problem in target formation by constructing a distance matrix, and to calculate the distribution of followers. In order to solve the problem of path conflict without error feedback, collision detection and collision avoidance are proposed, which effectively avoids motion failure. Experiment of transforming the line formation into wedge-shaped formation is simulated, and the result shows that the proposed formation transform method is feasible and can effectively improve the efficiency of formation transformation.

In modern times, the number of sensing image is increasing on explosive speed. Human’s ability on data analyzing and information accessing, however, has not caught up with this growth model, which requires our efforts to develop image mining technology, so that we can get what we need in short time. Classification and prediction method are two important contents on remote sensing image analyzing and information mining, as well as the focus of research. This paper has been written around the automation and intelligence of remote sensing image information obtaining and has done study on the remote sensing image data mining theory and technology. Besides, we have put forward an improved method on the Bayesian algorithm, having received a good effect.

In rescue missions, time is life, and only when the army arrives the first time around can risk to people’s lives and property be minimized. Therefore, not only does a troop’s transportation require reasonable dispatching but there is also a need to consider the time consumption problem of arriving at the location. Therefore, two factors – the number and time of transportation for disaster relief troops – are especially important. First, this study makes an in-depth analysis of the problem of troop dispatching in rescue and relief work, and proposes a network flow model of deploying troops thereof, thus making it a minimum cost maximum flow problem. Second, it defines the priority path to the existing minimum cost maximum flow algorithm; after joining the priority queue, the improved algorithm is more applicable to the study of a disaster relief troop’s transportation problem. Finally, experiments are done concretely on examples based on real life troop data. The results show that the model can effectively support the disaster relief troop’s transportation problem. The improved algorithm can effectively avoid small capacity and time-consuming deteriorated roads, and its time complexity is lowered from the original $$ O(n^{2} ) $$ to $$ O(n) $$ on a path selection judgment. The algorithm results can provide scientific reference for a disposal of contingency plans when danger occurs.

Micro-blog, as a kind of weak relationship network, strengthens the communication among the bloggers, and propagates instant information in the social network. With the explosive growth of information flow in social network, researchers have a growing realization that it is essential to accurately predict the cascading diffusion of a message, which is of paramount importance to applications like public opinion monitoring, viral marketing and outbreaks detection. Although there have been extensive previous works on diffusion prediction, what kind of factors affects the information diffusion most and how to predict the propagation process are the focusing issues all the time. This paper analyzes the information dissemination and forwarding mechanism in the social network. In particular, we extract main features from multiple dimensions including node attributes, message content characteristics and the topology relation between nodes. Based on these features, this paper proposed a cascades diffusion model to predict the propagation process. Besides, we quantitatively evaluated the weights of the features in the proposed model by a stochastic gradient descent algorithm. We evaluate the proposed method on Sina micro-blog dataset. The experimental results show that the proposed method outperforms the other common models in precise prediction.

Considering the issue of how to implement multi-target detection rightly in the complex environment of LFMCW (linear frequency modulated continuous wave) radar, a framework was proposed to analyze target detection based on width information in the frequency domain. The method involves three procedures. First, CFAR (constant false alarm rate) processor was implemented in frequency spectrum of beat frequency for the echo. Besides, a clustering algorithm was introduced to obtain width and amplitude information by calculate the continuous interval width of the frequency spectrum in positive frequency domain after the CFAR, Finally, multi-target detection by remove tiny spectrum cluster through width filtrating and spectrum line association utilize amplitude and width information in frequency domain. And a computer simulation was carried out. The result showed that the framework could effectively eliminate false target caused by clutter. The method could be used in target location and tracking signal processing in continuous wave radar system.

Android platform induces an open application development framework to attract more developers and promote larger market occupations at the same time. However, the open architecture also makes it easier to reverse engineering and application piracy. These result in the property loss for developers and companies, and increase the risks of mobile malicious code. Copyright protection for android application is thus of significant importance. Currently, many solutions for application copyright protection apply overload methods, assuming the availability of source code, which could be impractical for a large scale application protection. In this paper, we propose a lightweight copyright protection method for android application called DroidMark. The copyright is protected by text and space watermark based on semantics of xml and dex. Functional files are chosen as watermark carriers to increase watermark semi-fragileness and concealment. And the DroidMark can be accomplished without secret keys. Models and algorithms are proposed and analyzed all sidedly. The experiment results and analysis justified that DroidMark is secure and efficient.

In the field of rumor spreading, kill rumor is a very important concept. In the previous study of rumor spreading, only considering the rumor from the outside of the network, while ignoring the node itself has the discriminability. In this paper, a new CSER rumor propagation model is proposed and a kind of node with the ability of killing rumor is introduced in the model, that node is called rumor-killer. By means of complex network theory and mean-field method, we establish the differential equations of propagation dynamics. In this paper, the model is used to study the process of rumor spreading both in homogeneous network and heterogeneous network. Through theoretical analysis and experimental simulation, it is found that the node with the ability of killing rumor can mitigate the spreading rumor in the network. This conclusion provides us a new way to control the rumor spreading from the inside of network.