Search

Lip sync has emerged as a promising technique for generating mouth movements from audio signals. However, synthesizing a high-resolution and photorealistic virtual news anchor is still challenging. Lack of natural appearance, visual consistency, and processing efficiency are the main problems with existing methods. This paper presents a novel lip-sync framework specially designed for producing high-fidelity virtual news anchors. A pair of Temporal Convolutional Networks are used to learn the cross-modal sequential mapping from audio signals to mouth movements, followed by a neural rendering network that translates the synthetic facial map into a high-resolution and photorealistic appearance. This fully trainable framework provides end-to-end processing that outperforms traditional graphics-based methods in many low-delay applications. Experiments also show the framework has advantages over modern neural-based methods in both visual appearance and efficiency., Accepted by ICPR2020

Subspace projections can be used to suppress nonstationary interference, typically linear frequency modulation (LFM) interference, when the interference subspace is constructed correctly. When errors exist in the parameters used to construct the subspace, the interference suppression effect drops dramatically. To address this problem, the effect induced by the parameter estimation error is analyzed. On the basis of subspace error analysis, a method is proposed to remand the estimation error; the method finds the most accurate pair of parameters by comparing the inner products of the received signal and all the constructed interferences. When the frequency parameters are remanded, they can be used to construct the interference subspace; thus, through subspace projection, the interference in the received signal can be suppressed. In this study, this method was verified via simulation to be effective. By applying this method, the effect of subspace projection is promoted to a prominent degree, which makes it useful for LFM interference suppression.

The adaptive beamforming algorithm can be used to automatically form nulls to the direction of interferences, while the incident angle will variate because of the movement of the aerocraft, which make the Signal to Interference and Noise Ratio (SINR) drop dramatically. Take linear array as example, any movement of the aerocraft will result in the variation of the incident angle. So the incident angle is modeled as a variable in uniform distribution around its center. Through analyzing with optimization theory, the null broadening weight in explicit form is obtained, which includes the deduction of a so called average matrix and calculation of a Lagrange constant. The obtained weight can be used to confront the variation of incident angle. This method is certified in simulation to be effective. As applying the weight on the array, the SINR dropping caused by variation of incident angle is avoided, which improves the robustness of anti-jam algorithm.

Software testing is a key approach to guarantee the software quality. It is the objective unremittingly pursed by virtue of effective testing cases in the shortest time. The test suite optimization method is a NP-complete problem. Although some people apply k-means clustering algorithm to the test suite reduction, the algorithm is unstable and seldom considers the coverage rate of such test cases; as a result, it will waste many unnecessary testing time in redundant cases and always result in high cost. Therefore, this paper introduces k-medoids thought of the clustering algorithm and then proposes a method of parameter generation test suite characterized by cyclomatic complexity and code coverage rate. This method utilizes the greedy algorithm to process the streamlined test suite while guaranteeing the cases coverage rate and the error detection rate finally gain the minimal test suite. As indicated by the stimulation experimental results, our method features higher coverage rate with lower complexity under the streamlined test suite of the same quantity.

On the basis of Markov chain, this paper proposes a frequency and duration method for probabilistic load flow (PLF) analysis with wind power by improving the probability-frequency distribution function (PFDF) method. With this proposed method, the probability, frequency, and duration information of random output variables in the PLF problem can be efficiently computed by convolution and addition operation of PFDFs for random input variables. A numerical test on the modified IEEE-RTS79 grid verifies the effectiveness of the proposed method.

Forecasting results of regional photovoltaic (PV) installed capacity can provide important references for electric utilities and energy authorities. This paper proposes a three-step forecasting methodology of regional PV installed capacity. The first step is to study the relationship between regional PV installed capacity and a series of potential factors using co-integration analysis and Granger causality test; second, dimensionality reduction of influential factors is carried out by principal component analysis; in the last step, a grey support vector machine forecasting model is constructed based on the outcome of influential factors reduction. The example of Shanghai is given to illustrate the proposed methodology. Results show that factors such as electricity consumption, and generating cost of PV are closely related to PV installed capacity, nevertheless, sunshine duration and average temperature are not the Granger reasons for it; these findings are of great importance to the forecasting of 2016∼2030 PV installed capacity in Shanghai.

Background extraction is an important step in vehicle detection. In the actual scene, change of illumination will lead to a tremendous background change. It is necessary to update the background model reasonably and effectively as the illumination changes. In order to solve this problem, this paper proposes an adaptive ViBe background model. Firstly, two kinds of vehicle detection errors and their corresponding error function are defined. Then, according to the range of these two kinds of errors, a set of reasonable evaluation conditions are determined to adjust the unreasonable threshold value, which guarantees the adaptive updation of the background model. Experiments in real scenarios show that the adaptive ViBe background model has better vehicle detection accuracy than the mixed Gaussian model, the codebook model and the fixed threshold ViBe model.

WLAN can provide access to core network for mobile p hones, laptops and IoT devices. It becomes more necessary in some places such as office, shopping mall and stadium. Wireless devices need to connect with Access Point (AP) to obtain various kinds of service. Therefore, deployment of AP is important to guarantee QoS and user experience of the deployed WLAN. Most former works focus on signal coverage and cost of AP but neglect some problems in WLAN planning; for example, non-uniform user distribution, interference from existing WLAN system, and interference from other devices with same frequency. In order to solve these problems, we propose a WLAN planning method by using genetic algorithm to optimize placement and channel assignment of AP. We evaluate the network performance including delay and throughput by our method. The results show that WLAN planned by our method has better network performance by solving the problems.

Most existing works on video transmission over wireless multi-hop networks only consider routing as a network layer problem and have been decoupled from application. Traditionally, the multi-path routing protocol selects paths with lower hop counts and discards routes with higher hop counts, which are application-agnostic. However, the paths with minimum hop counts are not always the least delay ones. In this paper, we propose an application-centric multi-path selection and scheduling scheme for multi-description (MD) video transmission in wireless multi-hop networks, where the expected efficient transmission of video is adopted as the routing metric. The proposed routing scheme selects paths to transmit data with different priorities under the consideration of actual delay of the paths. In addition, the path switching scheme will be used to respond to the changes in the network environment. The simulation results demonstrate that the proposed application-centric approach can improve end-to-end delay and packet delivery ratio of I-frame video delivery and decrease distortion (in terms of PSNR) compared to conventional application-agnostic scheme.

Study in UAV aerial image quality assessment algorithm. Aiming at the characteristics of lack of reference images, have large amount of data and degradation type is mainly about blurry, we put forward a rapid and effective no-reference aerial image quality assessment algorithm based on the natural scene statistics. This algorithm reduces the computational complexity of traditional algorithms based on natural scene statistics by extracting the region of interest of human eye, utilizes the color information of aerial images effectively, and enhances the sensitivity of the algorithm to blurry by calculating color correlation blur metric. Experimental results show that the algorithm as a significant advantage in terms of computational complexity, and is more sensitive to the blurred image compared with the traditional evaluation algorithm. The proposed algorithm is more suitable for the UAV aerial image quality assessment.

In conventional multi-hop wireless networks towards IoT, network resources are static normally. As a result, if one network is overloaded, it cannot borrow any resources from other adjacent under loaded networks. The reason for this problem is that there is none of a uniform architecture and supporting techniques between these adjacent networks. In this paper, we propose a novel architecture and corresponding key techniques for multi-hop wireless networks. The proposed architecture are programmable and universal. By such proposal, all adjacent networks following the proposed architecture are managed by one control center. The network resources are optimally and dynamically managed among these networks to meet all application requirements.

With the development of the Internet, web service generates a large amount of log information, how to mine user preferred browsing paths from web log information is an important research area. Current researches mainly focus on the mining of user preferred browsing paths, however, they do not delve into the personalization of preferred paths and paths lack semantic information. To provide personalized preferred paths to fulfill users need, this paper proposes a innovative method of user preferred browsing path analysis based on vector space model. Firstly, path eigenvectors are adopted to denote the obtained preferred paths, and field eigenvectors are given by field experts. Secondly, the cosine similarity of path eigenvectors and field eigenvectors are computed. Thirdly, the preferred paths are partitioned into clusters according to the cosine similarity. Finally, the clusters are annotated using related fields. After clustering and annotation, the website can automatically recommend the related preferred paths for users according to the choice of users. Experiments show that it is accurate and scalable. It can be applied to optimize website or design personalized service.

A method is proposed to separate received LFM signals on underdetermined hypothesis. In which the sensors is less then the signals to be separated. As LFM signal belongs to nonstationary signal, the proposed method is based on Time Frequency Distribution (TFD) analysis. Through suppressing the cross term of the TFD, auto term become prominent. A rotation way utilizing the auto term points is presented, by which the TFD area of each LFM signal is classified. Using the STFD matrices corresponding to the auto term points, the Direction Of Arrival (DOA) of each LFM signal can be estimated individually. Through simulation, it is proved that the proposed method is effective.

Extraction of web page main text is one of the most efficient methods to improve search engine. In the traditional method, the extraction of the web page main text use the similarity of DOM sub-tree as a end condition for the DOM tree traversing, while its speed is unsatisfactory on such a complex web page structure. Thus, to raise the traverse speed and accuracy of DOM sub-tree effectively, we propose a method which is Semantics-based Extraction of Web page Main text.

In particle filter based vehicle tracking, the second order autoregression model and the fixed particle propagation radius are often used for particles sampling. This would produce certain errors and cause the particles to deviate gradually from the vehicle's true location in tracking. To resolve this problem, a modified state transition equation is built. In this equation, the vehicle's current location is estimated using the motion vector of its center block and the particle propagation radius is updated using kalman filter. Both improvements make the state transition equation more accurate. The experiment results show that the proposed method can decrease the particles' deviation and track vehicles more accurately than the particle filter using the second order autoregression model and the fixed particle propagation radius.

In our previous work [1], to improve search quality and user satisfaction by using the user's context of search we have developed a FOCS Model. In this Model, a context ontology is developed to record user's relevant context information and to help the semantic expansion of keywords. To make search results more relevant and personalized, similarity flooding algorithm is used to match context ontology with a faceted ontology, an ontology for annotating the target documents. However, similarity flooding algorithm cannot be used to calculate semantic data, so it is not a desirable method to solve the problem of context search well. Hence, we implemented a new Context Search in the model of IORCS. In this model, we integrate an inconsistent ontology reasoning method into our context search model to improve the accuracy of ontology matching step. First, we consider the ontologies we use in context search model as inconsistent ontologies. Second, to filter sub-ontology which the most similar to Search Ontology from related faceted ontologies, context reasoning function (CRF) is introduced to filter these inconsistent ontologies. Finally, context reasoning algorithm is used to implement ontology matching. The experimental results show the integration of inconsistent ontology reasoning into context search model can tackle the issue of Semantic Similarity Calculation better.

Vehicle recognition and classification is an important part of intelligent transportation system. Now, the technology of vehicle recognition has becoming a hot topic all over the world. A vehicle recognition algorithm based on Gabor wavelets transform and hidden Markov model (HMM) is proposed. A Gabor filters are applied on the vehicle images to construct a group of vectors called nodes, and then feature nodes are derived by using principal component analysis, which decrease the dimension of each node. The image including feature nodes is called Gabor-Vehicle. A set of images representing different instances of the same vehicle are used to train each HMM, and each individual in the database is represented by an optional HMM vehicle model. Experimental results show that the proposed algorithm has a high recognition rate with relatively low complexity.

This paper proposed an accurate shadow removal method for vehicle tracking. Firstly we detect and remove the shadow using optical gain based gradient analysis,in this process some parts of vehicle which are similar to shadow region in color space may be detected as shadow and removing these parts will leave some holes in the vehicle region. Then we fill these holes using the skeleton information of vehicle. Finally the shadow edge is eliminated by geometric scanning. Experimental results show that the shadow can be removed precisely using the proposed method.

Using the Internet for distributed graphics rendering has long been a dream of many image and animation producers. In this paper, the Alchemi and blender are combined to construct a grid-based graphics rendering system. A grid scheduling using Ant Colony Optimization (ACO) algorithm and Fuzzy Reputation (FR) Aggregation to improve the decision of scheduling is proposed, and a SAH-based kd-Tree used for the rendering process of blender as well. This method, which optimizes rendering with less time spent in the rendering process, in most cases does not affect the final quality of images obtained.

Identification of interactions between molecular features (e.g. mutation, gene expression change) and gross phenotypes in diseases and other biological processes is one of the important challenges in genomic research. Popular approaches such as GSEA are limited to hypothesis tests of bivariate association. However, a specific phenotype is often dependent upon multiple molecular features. It is thus worth considering all possible interactions jointly for a more precise and realistic representation of the cellular network. In this article, a semiparametric copula model is developed to jointly model genotypes, pathways and phenotypes to accomplish this object. A two-step procedure for reconstruction of the network is described. Simulation studies indicate that the method is effective and accurate for the network reconstruction. Application using NCI60 cancer cell line data identifies several subsets of molecular features that jointly perform as the predictors of clinical phenotypes. The copula model is expected to have a broad impact on biomedical research, ranging from cancer treatment to disease prevention.

The electrical properties of fish body will change during storage; this undoubtedly provides a new method to investigate the spoilage of fish. Based on virtual instrument technology and bio-impedance analysis method, a freshwater fish bio-impedance measurement system was built to investigate the bio-impedance characteristics of freshwater fish silver carp during storage. The result indicated that the impedance and phase decreased with post-mortem time, the accuracy of using impedance and phase angle index to determine fish freshness is 88.89% and 81.49. When taking the two parameters as the input, the TVB-N value as the output to build the ANN model to predict fish freshness, the model’s accuracy is 97.44%, which improves the accuracy of determination fish freshness and provides a new method to determine fish freshness.

Freshness is a key value for access fish quality; fish freshness could be determined by volatile compounds during storage. An electronic nose was constructed with virtual instrument and four commercial tin dioxide-based gas sensors array. Fresh water fish silver carp was stored at 4°C and tested by headspace method. A weighted piece of fish is introduced into a sensor's chamber and the signals of the sensors due to the fish emissions are recorded as function of time. Unsupervised method PCA (principal component analysis) was employed to discriminate among fresh, semi-fresh and deteriorate fish. PCR (principal component regression), PLS (partial least square) and BP-ANN (artificial neural network) method were used to build mathematic model to predict TVB-N (total volatile basic nitrogen). The results indicated that PCA method can discriminate fresh, semi-fresh and deteriorate fish, which indicated the electronic nose response correlated with the loss of fish freshness. The correlation coefficients of predicted and measured TVB-N value of PCR, PLS and BP-ANN model were 0.898, 0.922 and 0.976, which indicated that the BP-ANN method has the highest predicted accuracy for electronic nose.

With the enterprise applications being integrated into the pervasive computing environment, the traditional access control models based on the identity of the user and object have not been suitable to the new requirements of dynamic access authorization to enterprise resources. In this paper, we integrate business and pervasive computing requirements in a single model by extending the Task-Role Based Access model to provide dynamic context-aware access control for pervasive computing in enterprise environments. Unlike traditional approaches for access control, access decisions in will depend on the combination of the required credentials of users, their working environment and context of tasks. It provides a simple and efficient solution to integrate the business applications such as workflow systems into pervasive computing environment.