Abstract:Human motion capture data, as a new type of multimedia data, is widely used in many areas because of its high fidelity, but the expensive motion capture equipment yields the high cost of the use of motion capture data. Therefore, the technologies of motion capture data reuse become the effective means to solve the problem. However, the complex structure and characteristics of motion capture data make the motion capture data reuse challenging. Even it has been researched for many years, there are still many problems to be solved and more attentions and research efforts are needed. In this paper, in terms of the important technologies used in the process of motion capture data reuse, we give introductions on the research significance, difficulties, strategy and used models of current methods and so on. And we give a detailed description on some representative methods. Finally, we conclude the research advances of motion capture data reuse and discuss the possible directions for future works. This aims to cause the deep thinking of this field and provides a valuable reference for the future research.

Abstract:The stochastic optimization algorithm is one of the state-of-the-art methods for solving large scale machine learning problems, where the focus is on whether or not the optimal convergence rate is derived and the learning structure is ensured. So far, various kinds of stochastic optimization algorithms have been presented for solving the regularized loss problems. However, most of them only discuss the convergence in terms of the averaged output, and even the simplest sparsity cannot be preserved. In contrast to the averaged output, the individual solution can keep the sparsity very well, and its optimal convergence rate is extensively explored as an open problem. On the other hand, the commonly-used assumption about unbiased gradient in stochastic optimization often does not hold in practice. In such cases, an astonishing fact is that the bias in the convergence bound of accelerated algorithms will accumulate with the iteration, and this makes the accelerated algorithms inapplicable. In this paper, an overview of the state-of-the-art and existing problems about the stochastic first-order gradient methods is given, which includes the individual convergence rate, biased gradient and nonconvex problems. Based on it, some interesting problems for future research are indicated.

Abstract:Semantic processing is a key challenge in natural language processing. Abstract meaning representation (AMR) is a novel framework of representing the meaning of a sentence. Instead of a tree, it abstracts the meaning of a sentence into a rooted acyclic directed graph, which solves the argument sharing problem. Thus, the corpus construction and automated parsing of AMR become a heated research field. This paper introduces the AMR's basic concept, annotation guidelines, parsing algorithm and applications. Then we discuss the problems and shortcomings of the parsing algorithms by comparison experiments. We also introduce the development of Chinese AMR researches. At last, we discuss the potentials of AMR, which are fruitful for Chinese semantic processing.

Abstract:A new relay quantization scheme based on Gaussian sources is proposed, This model is applicable for the following system, the sensor can only simply send analog signals, and relays can provide distributed source encoding and channel encoding, the microwave radars and acoustic radars, etc. A theoretical analysis framework of the system is presented. The rate distortion function of the sensor network is established using chief executive officer(CEO) theory, and then we use the Shannon channel capacity theory to establish the connection between the sensor network and the digital communication network. The optimization design method of the system is proposed. Power allocation between the sensor network and the communication network is achieved to make the SNR performance reach the maximum under the condition of total power constraint. Theoretical analysis and simulation results show that the performance of the proposed method is much better than that of the analog amplify and forward. 

Abstract:Interval data (ID) is a kind of data which the attribute values are the interval. Aiming at the classification problem of interval data, a support vector machine classification model based on Gauss interval kernel (GIK_SVM) is proposed. In the method, the half-width factor is introduced which makes a compromise between the median and the half width of interval data. Then, the Gauss interval kernel is constructed to measure the similarity between two interval data. SVM model is applied to classify the samples.Experiment results on artificial and real datasets demonstrate that the proposed GIK_SVM has a better classification performance for interval data.

Abstract:The speaker clustering is an important process of speaker diarization, yet traditional method for hierarchical agglomerative clustering (HAC) with distance measurement based on Bayesian information criterion (BIC) can lead to the clustering error propagation. To solve this problem, step by step algorithm is proposed, when the minimum BIC distance between segments exceeds a predefined threshold, or the number of the categories on hierarchical clustering reaches a certain number. The current clustering result as the initial class center, and then variational Bayesian method will be exploited to tune the speaker segments among the categories iteratively. Finally, the number of speaker is determined according to the probabilistic linear discriminant analysis (PLDA) score threshold. Experiments on national institute of standards and technology (NIST) 08 summed test set show that this method improves the "class purity" and "speaker purity" compared with conventional algorithms. Moreover, performance of speaker diarization is relatively improved by 27.6%.

Abstract:A two-dimensional sine quadric surface filter is constructed and designed after analyzing the texture structure of fingerprint image and the comparability between the texture structure and two-dimensional sine quadric surface pattern. To reduce the effect of marginal noise on the performance of the filter and improve the enhancement capability of the filter, the paper adopts two-dimensional Gaussian function to modulate the two-dimensional sine quadric surface filter. Finally, the paper produces the fingerprint enhancement algorithm based on two-dimensional sine quadric surface filter modulated by Gaussian function. The grouping experimental results show that the proposed fingerprint enhancement algorithm can improve the quality of fingerprint image and reduce the effect of strong noise, such as the broked texture, the scar and the conglutination in the low-quality fingerprint image.

Abstract:Aiming at the existing problem of UMHexagonS algorithm in H.264, an improved fast motion estimation algorithm is proposed. The algorithm enhances the coding efficiency by optimizing the detection sequence of initial prediction motion vector(MV) according to the possibility of initial prediction MV becoming the best point, then an improved 5×5 spiral global search pattern is designed to reduce the search points of pattern, and an early termination technique for sub-macro block is proposed for further reducing the computation of motion estimation. The experimental results show that, with only negligible change of encoding performance, the improved algorithm can effectively reduce the motion estimation time and enhance the coding efficiency, and it can suit for video sequence of different motion intensities.

Abstract:A novel K-means algorithm of measurement partitioning is proposed to overcome the problem of distance partitioning algorithm in Gaussian mixture probability hypothesis density filter for extended target tracking. The number of the targets is estimated by maximum-likelihood estimator and then the estimates of the target number are used as the cluster number of K-means. An elliptical gate is introduced to remove the clutter measurements for depressing the influence of clusters. Simulation results show that the proposed algorithm reduces the computational complexity obviously, and obtains an improved performance.

Abstract:Energy-efficient beamforming design for multiuser is investigated in multicell systems to maximize the system energy efficiency subjected to some given quality of service demands and transmit power constraint. The non-convex optimization objective function in fractional form is firstly transformed into an optimization objective function in concave-convex fractional form, and the power allocation is realized by introducing some auxiliary variables and using jointly the fractional programming and the lower complexity convex approximation method. Then the beamforming optimization problem is transformed into a problem of minimizing the transmit power which can be solved with the second-order conic programming method. Numerical results illustrate that the proposed algorithm converges to a stable point within a limited number of iterations. It is also observed that the best spectral efficiency and energy efficiency can be simultaneously achieved by the proposed algorithm at low transmit power region. However, in high transmit power region the proposed algorithm outperforms obviously the traditional spectral efficiency maximization algorithm in terms of energy efficiency.

Abstract:The complexity of cascaded hydrothermal power system that its scheduling cannot be solved by the traditional method. Improved differential evolution algorithm is proposed and the scheduling problem are solved. In order to make full use of the information contained in the best individual of the population, variable weighting factor is applied to improve differential evolution algorithm, thereby the search ability of the algorithm is enhanced. The balance constraints of the scheduling problem is solved by heuristic strategy. The heuristic strategies based on priority list are devised to fully use thermal units with the lower average full-load cost and satisfy the dynamic balance constraints of power system. During the process of handling balance constraints, part of the individual’s value is changed, so that the search space is extended and the even better solutions are obtained. The simulation results show that the proposed approach can effectively solve the scheduling problem of cascaded hydrothermal power system.

Abstract:Image feature matching is a key link for the implementation of content-based image retrieval (CBIR), which mainly relies on the similarity measure between the features of two images. To improve the retrieval performance of CBIR, this paper proposes an effective similarity measure method—similarity measure based on k-nearest neighbors of images (SBkNN). In the proposed SBkNN method , the similarity between query image and retrieved image is obtained by calculating the probability for the two images belonging to the same semantic category (no matter what kind of semantic category), and the probability can be obtained by analyzing the distance between the two images and their k-nearest neighbors， respectively. Finally, the comparison between the proposed SBkNN method and traditional similarity measure is implemented on Corel5K dataset. Experimental results show that the proposed SBkNN method significantly improves the retrieval performance of CBIR.

Abstract:The key of pattern recognition is feature extraction. Fusion of feature is an important complement of feature extraction, and it has been proved to be important to improve discrimination. Here, the sparse representation method is studied by introducing sparse representation into a high dimensional feature space and utilizing kernel trick to make sparse representation in the space.The kernel sparse representation coefficients with kernel sparse representation are utilized, then kernel sparsity preserve projection (KSPP) subspace. Moreover KSPP is brought into canonical correlation analysis (CCA), then kernel sparsity preserve canonical correlation analysis (KSPCCA) is studied. The proposed algorithm is reliable and validated on the multiple feature database and face database.

Abstract:Interference factors such as seal cover, invoice crease and so on, cause noise adhesion in number area of some invoice, which would seriously lead to the invoice number segmentation error. Aiming at this problem, a noise adhesion area repairing algorithm is proposed. At the same time, according to the font structure and characteristics of ordinary invoice number, invoice number recognition algorithm based on characteristics of digital structure is proposed. Firstly, define number structure features, including four kinds of fill area, two kinds of number of passing through the character, and four kinds of hollow area, which constitute a 10-dimensional feature vector of the number to be identified. Then, match the feature vector with the template features in the standard template library, by obtaining the Euclidean distance, and regard the corresponding number with the minimum Euclidean distances as the last recognition result. The proposed method and printed number recognition method based on the improved left and right contour features are compared. Experimental results indicate that the proposed identification algorithm has higher accuracy, faster recognition speed and stronger robustness to noise.

Abstract:Identification and detection of the community structure is fundamental and important in the analysis of complex network. To detect community structure precisely, a new community detection algorithm based on EDA (Estimation of distribution algorithms) and field theory is proposed. By studying the instance relation of complex network and introducing the field theory, a community structure probability model is built. The proposed algorithm is illustrated and compared with GN (Girvan Newman) algorithm, genetic algorithm and heuristic algorithm by using classic real world networks. The result demonstrates the proposed algorithm is converge quickly and good practice.

Abstract:The signaling overheads of interference alignment (IA) obtaining global channel state information increase with the number of links. Grouping the links into clusters, within which the interference are processed by IA, becomes an effective method to reduce the overheads. Considering the fact of high computational complexity in the process of link partition, an alternative link partition algorithm based on minimum signal-to-interference-ratio (MinSIR) is proposed. Furthermore, when all the clusters simultaneously transmit in a single timeslot, the signal-to interference-plus-noise-ratio (SINR) at the receivers of several links was insufficient for successful transmission. To solve such a problem, the link scheduling problem was substituted by a novel cluster-based scheduling algorithm using hierarchical clustering. The theoretical analysis and simulation results show that the proposed link partition algorithm obviously reduced the computational complexity, and obtained superiors system throughput. Meanwhile, the cluster-based scheduling algorithm effectively improved the SINR at the receivers of links, which potentially supported the system decision of scheduling scheme for specified performance demand.

Abstract:Failing to identify multiword expression (MWE) may cause serious problems for many natural language processing (N LP) tasks. Because of lacking of Chinese MWE tagging corpus, a semi supervised method is used to extract Chinese MWE. DE-Tri-Training semi-supervised clustering algorithm uses supervised information in the beginning of the cluster, and obtains good results. The selection method of original cluster center based head word expansion and the consistency collaborative learning data depuration method based supervised information are proposed, which adds the supervised information into the mid and late steps of clustering, so that classifiers can use correct label information to train it. The contrast experiment show that the extraction results of Chinese multi-word expression using the improved DE-Tri-Training algorithm are better than that of using unimproved one. The effectiveness of the improved DE-Tri-Training algorithm is thus verified.

Abstract:Coreference resolution is a widely used technology to judge whether pronouns can match with the entity existing before in the text, which plays a crucial role in intelligent processing for massive text information on internet. A coreference resolution study, especially on the frequently-used Chinese personal pronouns, was carried out with the result of a developed algorithm with the combination of semantics and rules. Based on fundamental filtration rules, an improved mechanism specific to apposition was also adopted. To raise the accuracy of calculating the synonyms distances, the algorithm identified the associative-word of personal pronouns and selected antecedents based method for analyzing semantic relations and selecting high relevancy antecedent, which was realized with the aid of Tongyici Cilin and HowNet. Comparison experiments with different methods and experiments on the real corpus dataset were conducted, and results show that the presented method achieves higher validity and obvious improvement.

Abstract:A bidirectional enhanced diffusion filter image de-noising model is presented. The diffusion equation is firstly simplified and analyzed to establish bidirectional diffusion coefficient. Hence, the twoway process of smoothing and sharpening can be achieved by the model in the diffusion process, To further enhance the strength of the smoothing and sharpening, image enhancement is used to enhance the overall outline of the image using wavelet transform, thus weakening texture detail of the image. Then, the threshold will be designed and improved, and it will be automatically controlled by maximum image gray value and iterative times, which can retain the image edge and detail features. The proposed model is be simulated. The experimental result shows that the new model is ideal, and it can improve the performance of de-noising and the protection of edge. The texture detail information is satisfactory. The peak signal to noise ratio is promoted drastically. Therefore,the performance is better than classical algorithms.

Abstract:Based on existing software defect data, it is possible to improve the efficiency of software testing and reduce the test cost by establishing the classification model to predict the software modules. Most machine learning based defect prediction researches are based on two-way decision method. Since software defect prediction can be seen as a kind of cost-sensitive learning problem, and the software data has continuous values, this paper proposes a classification method based on neighborhood three-way decision-theoretic rough set model. For ambiguous testing modules, compared with two-way decision methods, this method makes a deferment decision to reduce the misclassification rate. Experimental results on NASA software datasets show that the proposed method can get a higher classification accuracy and a lower misclassification cost.

Abstract:Data sparseness severely affects the system performances of syntactic parsing, and syntactic structures are unities of syntactic forms and semantic contents. Based on the labeling of semantic information, a word clustering model and algorithm is proposed. And a head-driven statistical syntactic parsing model based on semantic category is established. The problem of data sparseness is successfully solved, and the system performances of syntactic parsing are obviously enhanced. Experiments are conducted for the head-driven statistical syntactic parsing model based on semantic category. It achieves 88.73% precision and 88.26% recall. F measure is improved 8.39% compared with the distinctive head-driven parsing model.

Abstract:To realize blind estimation of synchronization parameters of linearly digitally modulated signals in multipath fading channels and low signal-to-noise ratio (SNR), a joint algorithm for the estimation of carrier frequency offset, phase offset and symbol timing error based on cyclic cumulation is proposed. The mathematical relationships of cyclic cumulation and phase offset, symbol timing error are obtained through theoretical derivation. Based on this, the carrier frequency offset is roughly estimated firstly, then the quite precise estimation is obtained from the test of certain cycle frequency. Flinally, the phase offset and symbol timing error are estimated from the value of some certain cyclic cumulation. The algorithm do not rely on the color and distribution characteristics of the fading distortion or additive noise and is quite appropriate for complex condition that frequency offset, phase offset, time error and fading exist at the same time. Experimental results demonstrate the good performance of blind estimation of synchronization parameters in frequency selective fading channels under low SNR.

Abstract:Brain-computer interface (BCI) is a novel way for interaction between human and machine, which enables direct communication between the people′s mind and the computer. BCI system and its signal processing method are proposed based on pseudo random sequence modulated color visual stimulation and the chromatic transient visual evoke potential (CTVEP). According to the properties of human visual system, it has been demonstrated that chromatic visual stimulation is safer and more comfortable for various subjects. In the proposed BCI system, a number of spatially separated chromatic con-central rings will be simultaneously presented on and off, with on/off pattern each specified by a distinct sequence form 36 pseudo random Gold sequences. For the received data from occipital region, we have proposed a demodulation method based on the orthogonality among Gold sequences, and the subject focusing pattern is ultimately determined by using matched filtering. After experiments on 20 subjects, the identification accuracy of the proposed system was achieved, which has verified the validity of the system in translating the ideas of the subjects.

Abstract:The smileys with obvious sentiment orientation are easily annotated manually. But the annotations of the smileys with unobvious sentiment orientation are difficult to reach a consensus. A method of automatically determining the sentiment orientation of the microblog smileys with the seed words is proposed. The method automatically annotates the corpus smileys with obvious sentiment orientation using a few seed emotions. Then these smileys are used to generate the labeled smiley set (LSS). Moreover, a model is built based on the seed emotional words and LSS to determine the smileys with unobvious sentiment orientation. Experimental results show that the presented method is effective.