Search

In this paper, a novel image splicing detection and localization scheme is proposed based on the local feature descriptor which is learned by deep convolutional neural network (CNN). A two-branch CNN, which serves as an expressive local descriptor is presented and applied to automatically learn hierarchical representations from the input RGB color or grayscale test images. The first layer of the proposed CNN model is used to suppress the effects of image contents and extract the diverse and expressive residual features, which is deliberately designed for image splicing detection applications. In specific, the kernels of the first convolutional layer are initialized with an optimized combination of the 30 linear high-pass filters used in calculation of residual maps in spatial rich model (SRM), and is fine-tuned through a constrained learning strategy to retain the high-pass filtering properties for the learned kernels. Both the contrastive loss and cross entropy loss are utilized to jointly improve the generalization ability of the proposed CNN model. With the block-wise dense features for a test image extracted by the pre-trained CNN-based local descriptor, an effective feature fusion strategy, known as block pooling, is adopted to obtain the final discriminative features for image splicing detection with SVM. Based on the pre-trained CNN model, an image splicing localization scheme is further developed by incorporating the fully connected conditional random field (CRF). Extensive experimental results on several public datasets show that the proposed CNN based scheme outperforms some state-of-the-art methods not only in image splicing detection and localization performance, but also in robustness against JPEG compression.

The incompleteness of the knowledge base (KB) limits the performance of open domain question answering (QA).Represent the incomplete KB with graph attention network (GAT) and complement the incomplete KB by extra text achieves great success to boost the QA system when the KB is incomplete. In this paper, we propose a Graph-based KB and Text Fusion Interaction Network (GTFIN) to improve the performance of the incomplete QA system by utilizing the KB and text information.In GTFIN, to reduce the influence of the query-unrelated noisy information of GAT on final answer prediction, we first design a global-normalization graph attention network (GGAT) by determining the query-related edge weights from the global perspective, and then a coarse-to-fine text reader (CFReader) is proposed to both exploit the relation information and obtain the entity mention representation in the text to enhance the incomplete KB. We further incorporate a bi-attention mechanism to enhance the interaction between question and entity representation which could find more query-related entities for final answer prediction. On the widely used KBQA benchmark WebQSP, our model achieves state-of-the-art performance.

In order to obtain more accurate pose estimation, the visual-inertial odometry (VIO) system with outlier rejection and loop closure is proposed in this paper. Considering that feature matching is an important part in the front-end of the VIO system, its accuracy will affect the performance of the entire system. So we introduce an outlier rejection method of the grid-based motion statistics (GMS) algorithm to the VIO system. And for more robust feature correspondence and better camera pose estimation, we propose an improved GMS method to eliminate the mismatched points. Besides, we adopt the loop closure strategy to correct the cumulative error of the VIO system. Finally, we estimate the camera pose, velocity and IMU bias simultaneously by minimizing the loss function which contains reprojection error and IMU error. A large number of experiments on EuRoC demonstrate that the proposed method outperforms the advanced VIO system ROVIO and is comparable to the state-of-the-art VIO system OKVIS.

Sentiment computing brings some new application opportunities and technique challenges in artificial intelligence of the next generation, and it has become a fascinating research field. In this paper, the conception of sentiment computing with some core elements and feature vectors is defined, and some vital issues are proposed. Based on the theories mentioned above, the subjective content or objective content is classified by some special algorithms in the scenarios of single modal, such as text, image, audio and video data. Furthermore, how to merge these different kinds of data and to form the multimodal analysis methods for emotion detection is an important problem, and the fusion strategy is summarized in the paper. Finally, some trends about the sentiment cognition and sentiment generation are analyzed, which provides new ways for further research work.

In order to solve a series of problems about IT service management in Enterprise, such as lower efficiency and service quality, complexed business processes and communication patterns, huge static management data and dynamic control data, etc., an intelligence-awareness IT service management platform, which is based on the Cyber Physical System (CPS) for constructing smart Industry 4.0, is proposed to optimize the management and service process for IT resources in this paper. The whole IT resources of enterprise mainly include software, hardware and employee resources. How to aware and integrate all kinds of IT resources with intelligence method together for analyzing and providing better service is a big challenges, especially, the physical state-awareness about all hardware and network facilities, the business process-awareness about management software and business data, the knowledge-awareness and service among the employees in Enterprise. Based on these critical IT resources, a new smart ITSM platform is designed and programed, which is not only use CPS meta-model to build the state-awareness smart environment for IT service, but also integrate ITIL (Information Technology Infrastructure Library) meta-model as a standard management process to aware the change about the process and knowledge library. The application results show that the new CPS-Awareness ITSM platform promotes the efficiency and qualify of IT service and provides early-alarm in the management process of the whole life cycle of IT resources in enterprise.

In this paper, we present a new image forgery detection method based on deep learning technique, which utilizes a convolutional neural network (CNN) to automatically learn hierarchical representations from the input RGB color images. The proposed CNN is specifically designed for image splicing and copy-move detection applications. Rather than a random strategy, the weights at the first layer of our network are initialized with the basic high-pass filter set used in calculation of residual maps in spatial rich model (SRM), which serves as a regularizer to efficiently suppress the effect of image contents and capture the subtle artifacts introduced by the tampering operations. The pre-trained CNN is used as patch descriptor to extract dense features from the test images, and a feature fusion technique is then explored to obtain the final discriminative features for SVM classification. The experimental results on several public datasets show that the proposed CNN based model outperforms some state-of-the-art methods.

In order to forecast the price movement with the correlation between two different stocks, the model of Stock Social Network (SSN) is proposed to represent and analyze the intrinsic complex relationship. We choose 313 stocks from 9 industries to build an evolution model of SSN, which predicted that some stocks clusters are isolated and the nodes and edges in SSN are decreasing distinctly step by step with the change of threshold δ from 0.7, 0.75 and 0.8, respectively. Meanwhile, the coverage rate of nodes in SSN arrives 0.2076 at δ = 0.8, in reverse, the 79.24% nodes is trimmed during the process of evolution of SSN. Based on these results, we design a new portfolio strategy based on new index, named CSSNI, to optimize the asset pricing model. The results show that the ratio of return is 0.92666 based on the CSSNI, which is much better than the result by traditional strategy.

In order to forecast the price movement of stock with the correlated news events, an enhanced Topic-driven model with the positional weight of feature words and label of stocks, named LP-LDA model, is proposed to represent and analyze the intrinsic mechanism in financial market. The experiment results show that LP-LDA has a better performance than traditional LDA model. Especially, when the number of topics are increasing, the running time of LP-LDA model are 0.69s, 0.78 s and 1.15s at 100, 200 and 300 topics, respectively, which are better than LDA. Furthermore, Degree of Influence (DoI) is defined to describe the considerable influence about the news events on the price movement of certain stock, which provides a new mechanism to measure the fluctuating price. The experiment results shown that the coefficient of correlation between news topic and return rate of stock is 0.9137, which is much higher than other results of experiment.

Food image recognition is increasingly important for e-health applications. But this is a challenging topic due to the diversity of food, and color, light, view angles' effect on food image. Based on empirical and experimental explorations, we propose to use SIFT(Scale Invariant Feature Transform) and Gabor descriptors as food image features and KMeans algorithm for feature clustering. We also propose to use pervasive cloud computing paradigm to improve the performance of food image recognition due to the heavy computing requirement for large number of concurrent recognition requests. Evaluations show that the proposed approach can give acceptable recognition rate, and MapReduce programming can provide promising performance advantage compared to traditional client server approach.

Awareness of emotion status of people is fairly important for aged ones, the ones with sub-health status, and various patients in order to keep them in good mood. The emotion recognition at run time is intrinsically challenging due to its complexity nature. On the one hand, the awareness of human emotion should be achieved as non-intrusive as possible. On the other hand, the android smart phones on the market are increasingly popular which are equipped with various sensors that can be used to achieve the awareness of emotion status. In this paper, we propose an approach based on the heart beat rate and contents of user's talk, which are obtained from built-in camera and microphone on smart phones. We first classify anger, joy, normal, and sadness based on heart rates, then the emotion recognition is further improved by emotional key words in a talk. We have evaluated this approach in terms of recognition accuracy and power consumption found that the accuracy can achieve 84.7%.

Food image recognition is challenging due to the diversity of food, and color, light, view angles' effect on food image. The recognition process is also a computation heavy process. Therefore, We also propose to use pervasive cloud computing paradigm to improve the performance of food image recognition. Based on empirical and experimental explorations, we propose to use SIFT(Scale Invariant Feature Transform) and Gabor descriptors as food image features and KMeans algorithm for feature clustering. Evaluations show that the proposed approach can give acceptable recognition rate with good performance gains.

Pervasive computing is converging with cloud computing as with the development of smart cities. There arise the complexities from this convergence, like unprecedented large number of computing devices, flexibilities and adaptation of services for fitting personalization requirements from end users, expectations of powerful computing and storage capabilities of handhold devices, and so on. Therefore, a supporting infrastructure that can handle big data arisen from large number of devices, and can flexibly switch services at run time, is needed. Henceforth, in this paper, we propose an OSGi based pervasive cloud (OSGi-PC) infrastructure which can make use both of the cloud computing capabilities and component flexibilities from OSGi. We have evaluated the OSGi-PC in terms of performance and power consumption to show its usability.

This paper presents an ac/dc voltage doubler, where the output dc voltage is twice the input ac voltage amplitude. Two configurable power supply schemes are explored for the voltage doubler in vibrational energy harvesting applications. An `input-powered' power supply scheme eliminates standby power consumption of the circuit when the harvester is not active. Alternatively, a `cross-connected' power supply scheme reduces the minimum input voltage without additional circuit cost. Implemented in On Semi 3M-2P 0.5 μm CMOS technology, the circuit is first benchtop characterized with a signal generator and then also demonstrated with a vibrational energy harvester. A maximum power efficiency of 87% is achieved using the input-powered scheme for a 1.5 V ac input amplitude and 960 μW of output power. In the cross-connected scheme, the circuit is able to rectify input voltage amplitudes as low as 400 mV, below the transistor threshold voltage of 700 mV.

An etched implanted junction termination extension (JTE) is presented for high-voltage 4H-SiC JBS rectifiers. Unlike the conventional JTE structure, the proposed structure utilizes multiple etching steps to achieve the optimum JTE concentration range. Experimental and simulation results show that the JBS rectifier with etched JTE can improve the blocking performance compared to a conventional JTE structure and decrease the sensitivity of any possible variation in processing condition. The fabricated SiC JBS rectifier showed the forward on-state voltage characteristic is 1.3V at room temperature and the blocking voltage of 1.2kV.

This paper presents an input-powered energy-harvesting interface circuit that eliminates standby power consumption by automatically shutting down when the ac input voltage amplitude is too low for successful energy reclamation. The interface comprises an input-powered ac/dc converter and an input-powered dc/dc boost converter that are fabricated in ON Semi 3M-2P 0.5 µm CMOS process. The interface circuit starts up when the ac input amplitude exceeds 1 V and supplies a regulated dc output up to 3 V. When the input amplitude drops below 600 mV, the interface automatically enters standby mode and consumes no power. The system achieves a maximum efficiency of 60% with 1.5 V ac input amplitude and 3 V regulated dc output, delivering 3.9 mW of output power. The interface also functions properly in tests with an electrodynamic (magnetic) vibrational energy harvester.

By the strategy of exerting advantage and avoiding disadvantage respectively, hybrid GEP (Gene Expression Programming) algorithm includes many advantages, which can effectively improve the efficiency of GEP and offer more effective solution to the difficult problems of technology and engineering fields. This paper proposes GEP-PTS algorithm, which combines simple GEP and PTS. In GEP-PTS, we propose Parallel Taboo Search (PTS) based on simple taboo search to conduct local search. Extensive experiments show that the accuracy of function model found by GEP-PTS is improved 4.19%-7.93% compared with simple GEP.

Due to the varying population density, some ISLs of satellite networks are congested while others are underutilized. To satisfy the QoS requirements of multimedia applications, satellite routing protocols should consider the issue of improving the utilization of network resource and providing better QoS guarantees such guaranteed end-to-end delay bound as new call blocking probability (CBP) and handover call blocking probability (HBP). This paper proposes a QoS routing protocol for LEO satellite networks using genetic algorithm. This protocol is called as Genetic Satellite Routing Protocol (GSRP). Some key factors including Fitness Function, Termination Function, and Routing Table, are newly taken into account. Simulation results show that GSRP is to achieve (1) guaranteed end-to-end delay bound with less CBP and HBP, and (2) more effective than traditional algorithms for load balancing.

This paper presents a novel model of artificial immune system for network intrusion detection. It starts by introducing rationale and techniques of natural immune system and IS-based intrusion detection system. A multi-shape-gene model is presented with its physical and logic architecture, as well as technology of the generation and evolution of genes libraries. Finally the negative selection algorithm is introduced.