Skip to main content

Über dieses Buch

Proceedings of the 2013 Chinese Intelligent Automation Conference presents selected research papers from the CIAC’13, held in Yangzhou, China. The topics include e.g. adaptive control, fuzzy control, neural network based control, knowledge based control, hybrid intelligent control, learning control, evolutionary mechanism based control, multi-sensor integration, failure diagnosis, and reconfigurable control. Engineers and researchers from academia, industry, and government can gain an inside view of new solutions combining ideas from multiple disciplines in the field of intelligent automation.

Zengqi Sun and Zhidong Deng are professors at the Department of Computer Science, Tsinghua University, China.



Chapter 1. 3D Point Cloud Based Hybrid Maps Reconstruction for Indoor Environments

In this article we investigate the problem of constructing a useful 3D hybrid map for both human being and service robots in the indoor environments. The objects in our laboratory include different tables, shelves, and pillar, which are of great importance for indoor service robot. We detail the components of our map building system and explain the essential techniques. The environment is detected in 3D point clouds, after sophisticated methods operating on point cloud data removing noise points and down sampling the data, we segment the data into different clusters, estimate the posture for clusters that can be recognized from library and replace it with VRML model we built in advance, then reconstruct surface for which cannot be recognized. Finally the preliminary hybrid maps are represented with the form of point cloud, VRML model and triangular meshes in 3DMapEidtor.

Biao Zhang, Qixin Cao

Chapter 2. Efficient Discriminative K-SVD for Facial Expression Recognition

Dictionary learning has attracted growing intention for its prominent performance in many computer vision applications including facial expression recognition (FER). Discriminative K-SVD (D-KSVD) is one of conventional dictionary learning methods, which can effectively unify dictionary learning and classifier. However, the computation is huge when applying D-KSVD directly on Gabor features which has high dimension. To tackle this problem, we employ random projection on Gabor features and then put the reduced features into D-KSVD schema to obtain sparse representation and dictionary. To evaluate the performance, we implement the proposed method for FER on JAFFE database. We also employ support vector machine (SVM) on the sparse codes for FER. Experimental results show that the computation is reduced a lot with little performance lost.

Weifeng Liu, Caifeng Song, Yanjiang Wang

Chapter 3. An Improved RANSAC Algorithm of Color Image Stitching

Image mosaic is a technique being used to stitch multiple images together to form a stitched image with higher resolution and large field of view. This paper presents an improved RANSAC algorithm of color image mosaic. Image mosaic mainly comprises two steps, namely image registration and image fusion. This algorithm comprises main modules, such as Harris corner detection, NCC rough matching, improved RANSAC exaction, estimating the projection transformation matrix, projection transformation, image smoothing. The improved RANSAC for feature points exaction based on statistical regularities can greatly support the real time of the algorithm. The experimental results show that the algorithm presented is fast and effective.

Weijie Huang, Xiaowei Han

Chapter 4. Target Tracking Algorithm Based on Visual Perception Mechanism

A method based on visual perception mechanism is proposed for solving the problem of target tracking. The tracking of target can be achieved in stability. In this paper, the algorithm use neural responses as the visual features. Firstly, the receptive field of cells in primary visual cortex is obtained from natural images. Then the neurons response of background image and video image sequences can be received and calculated the difference, and the difference is compared with dynamic threshold, the target can be detected in this way. Finally, the target tracking can be realized by iterative. Many categories experiment results show that this method improve accuracy and robustness of the tracking algorithm in condition of time-real.

Peng Lu, Shilei Huang, Chi Liu, Daoren Yuan, Yafei Lou

Chapter 5. Positioning Tropical Cyclone Center in a Single Satellite Image Using Vector Field Analysis

Tropical cyclone (TC) center location or eye fix is the first and inevitable procedure in TC forecasting, but mostly done manually in practice. Here we present a novel and objective method to locate TC center by compensating conventional pattern matching and vector field analysis methods together. The experimental results show satisfactory results in comparison with the best track data from the China Meteorological Administration especially when the spiral patterns of TC are well structured. Our method being introduced, TC forecasters can fix TC eye objectively and easily from data of all kinds of meteorological satellites.

Jinfeng Yang, Haoren Wang

Chapter 6. Target Extraction Study on the Vision System of Apple Picking Robot

This paper studied the color images of mature apple in natural environment, there is a general distinction between mature apple’s color and the background. Apple and background distributed in different area of color space. According to this characteristic, this paper proposed an object extraction algorithm based on sample color space. First we construct the sample color space in L*a*b* space by using apple samples’ image and using mathematical morphology to optimized it. Then recognised the apple target according to the sample color space. For the small target of depth of field and serious keep out targets, we make them into binarized images and use morphology structure element again to processing them. At last we got the ideal segmentation effect with high recognition rate.

XinWen Cheng, XueQiang Shi

Chapter 7. The Research on the Method of Traffic Area Dynamic Division and Optimization

Based on the Temporal and spatial variation characteristics of traffic flow, this paper puts forward the traffic area dynamic division method and how to realize the merger and separation of nodes in the control area, as well as inter-regional automatic combination and split. First, according to the traffic relevance and traffic similarity between adjacent intersections in the traffic area, the application of clustering analysis method, the traffic area is divided into some dynamic control zones. Then, application node shrinkage method is to determine the key nodes of the control zone, and relying on the fuzzy cognitive map is to analyze the direct and indirect influence of the key nodes and the other nodes in the control zone. Finally, according to the traffic flow state changes forms: basic remains the same, crowded diffusion and crowded subsided, respectively, is corresponding to the dynamic changes of the control zones: remains unchanged, expansion combination and shrinkage decomposition, in order to achieve the merger and separation of the nodes in the control zone, as well as inter-regional automatic combination and split.

Yirong Guo, Baotian Dong, Lei Wu

Chapter 8. Multiclass Vehicle Detection Based on Learning Method

This paper presents a real time vehicle detection framework using learning method. This framework combines offline multiclass support vector machine and online boosting method for vehicle detection. Compare to tradition approaches, the proposed method can robust deal with multiclass vehicles and unfamiliar environment. Experiments with the city vehicle database are used to evaluate this method. The experimental results demonstrate the consistent robustness and efficiency of the proposed method.

Zhiming Qian, Jiakuan Yang, Lianxin Duan

Chapter 9. An Object-Level Approach Improved by Quadtree to Dynamic Monitoring of Mining Area Expansion

An object-level approach improved by quadtree to dynamic monitoring of mining area expansion is proposed. In order to improve the efficiency and quality of objects acquired from high spatial resolution remote sensing image, multi-scale segmentation combined with quadtree segmentation is used to obtain objects of multitemporal remote sensing images; Then object-oriented image analysis method which takes into account the spatial relationship between ground objects is used in multitemporal remote sensing images to extract mining information respectively; Finally, overlay is use in mining areas extraction respectively, and Inter-erase operation is used to obtain result of mining expansion. Experiments are carried out in remote sensing images from a certain phosphate area of Anning, and the results prove that method was proposed in this paper is feasible and effective.

Liang Huang, Yuanmin Fang, Xiaoqing Zuo, Xueqin Yu

Chapter 10. Heuristic Optimization Algorithms for Solving MRMPT

Multi-vehicle ride matching problem with transfers (MRMPT) studies how to match passengers as much as possible to vehicles at the basis of MRMP which has no transfer mechanism. The passengers in this problem are classified according to their needs. The first-level passengers have been matched in the basic MRMP. And what we are solving is matching the second-level passengers in MRMPT. To improve the riding rate, a heuristic algorithm based on ant optimization for the MRMPT is proposed. The algorithm is divided into three steps: Finding starting and ending sets; optimizing the routes by ant optimization; Triming the vehicle routes. Simulation experiment shows that the algorithm can achieve riding rate of 80 %. The results show clealy that the algorithm can find the matching routes with high efficiency and low cost.

Chunhua Meng, Hongguo Wang, Zengzhen Shao, Yanhui Ding

Chapter 11. A Delay-Based Analysis of Multiple Bottleneck Links of End-to-End Paths in the Internet

Measurement and analysis of bottleneck links play an important role in improving the network quality of service (QoS) and preventing network attacks in the Internet. Existing methods usually treat the link with the smallest available bandwidth or the largest delay as the bottleneck link, without considering multiple bottleneck links in an end-to-end path. In this paper, we propose a new approach to measure and analyze bottleneck links based on path delay. We design a parallel active measurement framework to measure the path delays of many destinations simultaneously. Then an algorithm to identify multiple bottleneck links is proposed using the Ward data clustering method. Experiments are conducted to test the algorithm by measuring 10 different destinations in the Internet for 14 days. Using the proposed approach, we have found that bottleneck links are mainly few constant links which are in the intermediate of end-to-end paths or near the destinations. Furthermore, the results have shown that the number of intra-domain bottleneck links takes a large portion in most cases, which hints that the performance of end-to-end paths may be greatly influenced by iBGP routing. Besides, the results have also demonstrated that the intercontinental links in an anonymous system (AS) incline to be bottleneck links in end-to-end paths.

Jingang Liu, Wei Peng, Yonglei Yang, Zhijian Huang

Chapter 12. Hand Segmentation Based on Skin Tone and Motion Detection with Complex Backgrounds

Hand Segmentation is the first problem need to be solved in hand recognition system. Currently, most hand gesture recognition system is based on simple background, or requests the recognizer on glove in special color, which gives human–computer interaction some restrictions. This paper researches the gesture segmentation technology based on complex backgrounds, and gives a method combined with skin tone detection and motion detection. By experiments on the images captured by home security robot, this method can get accurate hand segmentation of all the images. This paper lays the foundation of gesture recognition on the home security robots.

Xintao Li, Can Tang, Chun Gong, Sheng Cheng, Jianwei Zhang

Chapter 13. Local Reconstruction and Dissimilarity Preserving Semi-supervised Dimensionality Reduction

In this paper, a semi-supervised dimensionality reduction algorithm for feature extraction, named LRDPSSDR, is proposed by combining local reconstruction with dissimilarity preserving. It focuses on local and global structure based on labeled and unlabeled samples in learning process. It sets the edge weights of adjacency graph by minimizing the local reconstruction error and preserves local geometric structure of samples. Besides, the dissimilarity between samples is represented by maximizing global scatter matrix so that the global manifold structure can be preserved well. Comprehensive comparison and extensive experiments demonstrate the effectiveness of LRDPSSDR.

Feng Li, Zhengqun Wang, Zhongxia Zhou, Wei Xue

Chapter 14. Fusion of Gray and Grads Invariant Moments for Feather Quill Crease Recognition

In order to overcome the non-crease misjudgment of feather quill, a novel decision fusion algorithm is proposed. An improved Radon transformation is used to extract moment invariants of gray and grads dual-mode of target region and singular value decomposition is provided here to obtain feature vectors, respectively; then creases recognition is performed according to feature vectors of the dual-mode. Finally, the final recognition result of the system is achieved by the fusion of recognition results of the dual-mode at the decision level. Experimental results show that this new method can overcome the limitations of single-modal and reduce the misjudgment of non-crease effectively.

Hongwei Yue, Renhuang Wang, Jinghua Zhang, Zuihong He

Chapter 15. A New Vision Inspired Clustering Approach

In this paper, a new clustering approach by simulating human vision process is presented. Human is good at detecting and segmenting objects from the background, even when these objects have not been seen before, which are clustering activities in fact. Since human vision shows good potential in clustering, it inspires us that reproducing the mechanism of human vision may be a good way of data clustering. Following this idea, we present a new clustering approach by reproducing the three functional levels of human vision. Numeric examples show that our approach is feasible, computationally stable, suitable to discover arbitrarily shaped clusters, and insensitive to noises.

Dequan Jin, Zhili Huang

Chapter 16. Multiclass Vehicle Tracking Based on Local Feature

This paper presents a real time multiclass vehicle tracking method. The method uses a combination of machine learning and feature analysis to track the vehicles on the road. Multiclass SVM are trained using train samples to achieve detection and classification of vehicles in video sequences of traffic scenes. The detection results provide the system used for tracking. Each class vehicle is tracked by SIFT method. Experimental results from highway scenes are provided which demonstrate the effectiveness of the method.

Zhiming Qian, Jiakuan Yang, Lianxin Duan

Chapter 17. Simulating Virtual Plants Based on Genetic Algorithm and L-Systems

As an effective technique to dynamically simulate the morphological development of plants, L-system is widely used in the field of plant modeling and visualization. It is a key step to obtain the axiom, the productions, and theirs parameters before simulating a special target. Usually, it is difficult to get those parameters and rules. In order to avoid the blindness and low efficiency in the modeling process, we proposed an automatic algorithm to simulate the images of virtual plant based on Genetic Algorithm and L-systems in this article. The relative techniques of our methods, such as the encoded mode of genetic individuals, the design of the primary population, the design strategies of genetic operations, and the design of evaluation function of fitness are introduced in details. The results demonstrate the algorithm’s capability of modeling various plants and ease of use by novice users.

Weilong Ding, Chen Hu, Yuanwei Zhu

Chapter 18. Shrinkage Common Spatial Pattern for Feature Extraction in Brain-Computer Interface

Common spatial pattern (CSP) has been one of the most popular methods for EEG feature extraction in brain-computer interface (BCI) application. Although the CSP usually provides good discriminant features for classification, it is also known to be sensitive to overfitting and noise. This study introduces a shrinkage technique to regularize estimation of the covariance matrices in the CSP and hence a novel shrinkage CSP (SCSP) method, which could effectively alleviate the effects of small training sample size and unbalanced data on classification. The proposed SCSP is validated on feature extraction of P300 that has been widely adopted for BCI development. Classification accuracies are evaluated by using linear discriminant analysis (LDA) with experimental EEG data from seven subjects. The results indicate that the proposed SCSP extracts more effective features that yield higher classification accuracy than that by the traditional CSP.

Yu Zhang, Jing Jin, Bei Wang, Xingyu Wang

Chapter 19. Detection of Overtaking Vehicles in a Highway

In this paper, some challenges of detection of overtaking cars in a highway are reviewed. Based on this analysis, we propose an overtaking vehicle detection method based on the computation of a symmetry function. Our approach defines a new symmetry function. Although the near distant vehicles usually do not show strong symmetry, we can also use this function to detect it locations. The proposed approach has been evaluated using some practical traffic scenes.

Song Pan, Huaping Liu

Chapter 20. A Generalized Gamma Distributed CFAR Algorithm for Layover and Shadow Detection in InSAR Images

In this paper, a novel CFAR algorithm is proposed for layover and shadow detection in Interferometric synthetic aperture radar (InSAR) images. Firstly, the generalized gamma distribution (GГD) is employed for statistical modeling of the InSAR image. Moreover, a CFAR algorithm for detecting both the layover and shadow is proposed, of which the analytical expressions of two thresholds are presented. Finally, the effectiveness of the GГD and the proposed CFAR algorithm are validated with a real InSAR image.

Xianxiang Qin, Huanxin Zou, Shilin Zhou, Yun Ren

Chapter 21. Error Concealment via Best Matching Block Selection Strategy

In this paper, we propose a new error concealment method based on best matching block selection strategy, which is implemented under a Bayesian probabilistic framework. Best selected image block is forwarded to a block-based bilateral filter for restoring missing pixels, while operates in block wise manner. During the Bayesian framework, a non-local spatial correlation is introduced in block selection. Meanwhile, a Markov Random Field Prior is built as an image prior model. Finally, experimental results show favorable results of proposed algorithm against traditional methods.

Chen Yao, Lijuan Hong, Yunfei Cheng

Chapter 22. Preliminary Evaluation of Classification Complexity Measures on Imbalanced Data

Classification complexity measures play an important role in classifier selection and are primarily designed for balanced data. Focusing on binary classification, this paper proposes a novel methodology to evaluate their validity on imbalanced data. The twelve complexity measures composed by Ho are evaluated on synthetic imbalanced data sets with various probability distributions, various boundary shapes and various data skewness. The experimental results demonstrate that most of the complexity measures are statistically changeable as data skewness varies. They need to be revised and improved for imbalanced data.

Yan Xing, Hao Cai, Yanguang Cai, Ole Hejlesen, Egon Toft

Chapter 23. A Real-Time Tracking Algorithm Based on Gray Distribution and Distance Kernel Space

The application of the traditional Camshift algorithm, which exhibits a good tracking performance in case of the obvious color characters, meanwhile, is limited in the target tracking in the color space. A fast tracking algorithm based on gray value distribution and distance kernel space is proposed. A 1.5D gray histogram method is designed to describe the model of moving object, which improves the reduction of computation for the back projection and real-time tracking performance. Moreover, a distance kernel function, describing the object weights, is constructed so as to handle the background disturbance and occlusion problem. Experiment results demonstrate the efficiency of proposed algorithm, that it can achieve a fast object tracking and resist background disturbance in some level.

Weixing Li, Yating Xiao, Feng Pan, Kai Zhou

Chapter 24. Laplacian Regularized D-Optimal Design for Remote Sensing Image Classification

Obtaining training sample for remote sensing image classification is time consuming and expensive especially for relatively inaccessible locations. Therefore, determining which unlabeled samples would be the most informative if they were labeled and used as training samples is the most delicate phase. Particularly, we consider the problem of active learning in remote sensing image classification. However, Classical optimal experimental design approaches are based on least square errors over the labeled samples only. They fail to take into account the unlabeled samples. In this paper, a manifold learning technique which is performed in the sample space by using graph Laplacian is applied to reflect the underlying geometry of the sample. By minimizing the least square error with respect to the optimal classifier, we can select the most representative and discriminative sample for labeling. The effectiveness of the proposed method is evaluated by comparing it with other active learning techniques existing in the literature. Experimental results on data set confirmed the effectiveness of the proposed technique.

Kang Liu, Xu Qian

Chapter 25. Speaker Tracking Based on Audio-Visual Fusion with Unknown Noise

In order to meet the high precision, strong robust demands of speaker tracking system, this paper proposed a new Particle filter algorithm with unknown noise statistic characteristics. The proposed algorithm estimate and correct the statistic characteristics of the unknown noise on-line by improved Sage-Husa estimator, and produce optimal distribution function with unscented Kalman filter. Finally, it realized speaker tracking problem based on audio-visual fusion in the framework of the new algorithm. Experiment results show that the method proposed in this paper has enhanced the accuracy and robustness of speaker tracking system.

Jie Cao, Jun Li, Wei Li

Chapter 26. The Analysis of Epidemic Disease Propagation in Competition Environment

An autonomous SIRS epidemic model of two competitive species is established in this paper, in which One kind of disease can survive and have the chance of cross-infection, not only the disease of the cross-infection but also the additional disease death rates are considered in this model, through the analysis of this kind of model, we can gain the threshold value condition of the stability of equilibrium. Even more it can see the global stability of the model through simulation number value.

Mingsheng Hu, Suimin Jia, Qiaoling Chen, Zhijuan Jia, Liu Hong

Chapter 27. An Entity Answer Ranking Method Based on MLNs

For the characteristics of factoid and list answers in domain Q&A system, we built a ranking model combined with multiple features of domain entity answers based on MLNs. This method uses predicate formulas to describe the relevant features of the questions–candidate answers and the answers–knowledge base, merging these features into Markov Logic Network, and then adopting discriminant training learning algorithm to learn the weights of feature parameters, which can give different weights according to the relevance of different features, finally it use MC-SAT algorithm reasoning to get the relevance of questions and answers, realize the answer ranking. Experiments show that the proposed method can greatly improve the answer precision and recall rates compared with other methods.

Fangqiong Chen, Zhengtao Yu, Jianyi Guo, Tao Shen, Yantuan Xian

Chapter 28. A Chinese Expert Name Disambiguation Approach Based on Spectral Clustering with the Expert Page-Associated Relationships

Aimed at the problems of Chinese experts’ name repetition and representation diversity, a Chinese expert name disambiguation approach based on spectral clustering with the expert page-associated relationships is proposed. Firstly, the TF-IDF algorithm is used to calculate the word-based feature weights, and then the cosine similarity algorithm is employed to compute the similarity between the evidence-pages to obtain the initial similarity matrix of expert evidence-pages. Secondly, the expert page-associated relationship features are taken as the semi-supervised constraint information to correct the initial similarity matrix, and next the spectral clustering-based method is used to build expert disambiguation model. Finally, taking the contrast experiments on Chinese expert evidence-page corpus of manually labeled, the result shows that the semi-supervised spectral clustering on Chinese experts’ name disambiguation method with the expert page-associated relationships than that without the associated constraint information, the F-value has an average increase of 9.02 %.

Wei Tian, Tao Shen, Zhengtao Yu, Jianyi Guo, Yantuan Xian

Chapter 29. The Hierarchical Heterogeneous of Parallel Computing Model Based on Method Library

This paper puts forward a novel hierarchical heterogeneous of parallel computing model that based on method library. In the original models, although the models provide a model algorithm language for usage, the developers still need to rewrite the methods when calling and the usage is apparently very complex for the developers, demanding the developers’ programming skill. In this regard, this paper presents a new hierarchical heterogeneous of parallel computing model which is based on the method library as well as its management system. While developers use the methods, they should only know the parameters of the methods called without having to know the methods of the preparation process, which can help the developers call the method library methods easily to facilitate the preparation of the algorithm program. From experiments results based on the method library, conclusions are drawn. Experiments show that the introduction of the method libraries and its management systems can not only reduce the difficulty of developers, but also reduce compile time and accelerate the operation of the program.

Jibing Duan, Xiaopeng Ji, Jinye Dou, Zhiqiang Wei

Chapter 30. Research on Weakly-Supervised Entity Relation Extraction of Specific Domain Based on Entropy Minimization

There are two major issues of automatic entity relation extraction: human intervention and difficulty in labeling corpus. For these two problems, combined with the characteristics of the tourism domain, this paper adopts a weakly-supervised extraction method of entity relation based on entropy minimization. This method firstly extracts the characteristic words by the idea of scalar clustering with small-scale stratified marked instances, and constructs the initial classifier with maximum entropy machine learning algorithm. Then use the initial classifier of certain accuracy to classify the unlabeled instances, and add the instances of the minimum information entropy to the training corpus set to continually expand the scale of training data set. Finally, repeat the above iterative process until the performance of classifier is to be stabilized, and then a final extraction classifier of entity relation in specific domain will be constructed. Experiments performed on the corpus of tourism domain show that, not only can this method reduce the dependence of entity relation extraction on manual intervention, but it could effectively improve the performance of entity relation extraction, the F value of which is up to 63.69 %.

Jun Zhao, Jianyi Guo, Zhengtao Yu, Peng Chen, Cunli Mao

Chapter 31. Using Fast Sampling-Insensitive Stereo Matching for 2-D Face Recognition

In this paper, we propose using sampling-insensitive stereo matching for 2-D face recognition. We don’t perform 3-D reconstruction but define a measure of the similarity of two faces. Then we match one 2-D query image to one 2-D gallery image using the measure for face recognition. We show that this method is not only robust to pose variations but also faster than other stereo matching methods. The proposed approach has been tested on the CMU PIE and ORL data set and demonstrates superior performance compared to existing methods in real-world situations including changes in pose and illumination.

Rui Liu, Longfei Cui, Wenke Zhang, Ming Zhu

Chapter 32. The Detection Method of Printed Registration Deviations Based on Machine Vision

In order to meet the actual requirements of judging four-color printed registration deviations quickly and accurately, this article describes a new detection method based on machine vision and the method has been designed involving three processes: removing interferential image by a corresponding region character separation (RCS) algorithm; detecting the edge of registration mark by interpolation subpixel algorithm and use weighted markov chain to calibrate the detection. The experiment indicates that the speed and accuracy with this method have greatly improved, and even with noise interference, this method can detect deviation quickly and accurately, superior to the traditional detection method.

Kailong Liu, Minrui Fei, Wenju Zhou, Haikuan Wang

Chapter 33. Community Discovering Based on Central Nodes of Social Networks

Based on the new concept of central network and the node similarity definition, a fast algorithm is proposed for discovering community structures in social networks. Firstly, start from the maximum degree node, then the two nodes with the maximum number of shared neighbors are taken as the initial community. Next, a neighboring node is judged to push into the initial community according to the appearance frequency of the community it belongs to. Finally, the above step is repeated until all the nodes are classified to the proper communities. The experimental results on two real-world networks demonstrate that the proposed algorithm is able to discover community structure from a given network efficiently and accurately without specifying the community number. The algorithm has a time complexity of only O(n).

Ping Fang, Fenglong Shi, Yang Chen, Wanchun Gao

Chapter 34. An Improved Force-Directed Algorithm Based on Emergence for Visualizing Complex Network

Visualization of complex network is one of the most important and difficult issues in complexity science. Most current visualization algorithms are based on drawing aesthetically and it’s difficult for them to find the structure information of complex network. To solve this problem, the Fruchterman-Reingold (FR) algorithm based on the force-directed layout is studied in this paper, which is most suitable for the visualization of complex network. From the emergent characteristic of complex network, an improved adaptive FR algorithm is proposed to reduce dependence on parameters in the FR algorithm. In the improved algorithm, the impact of the clustering coefficient on the attraction–repulsion between vertices is considered, and the clustering coefficient is thought to be a determinative indicator for emergence. Then, with the attraction–repulsion the topological characteristic of complex network is visualized. Experiments show that the improved algorithm makes the observation of the structure of complex network much easier. In addition, the improved algorithm displays superior stability and adaptability during experiments.

Hongbo Li, Wenjing Geng, Yu Wu, Xian Wang

Chapter 35. Exploring Efficient Communication in Interactive Dynamic Influence Diagrams

Interactive Dynamic Influence Diagrams (I-DIDs) provide an efficient method for representing multiagent sequential decision problem. By extending I-DIDs with communication, agents are able to exchange their information to learn more about the world. Note that communication is not free, agents should decide whether to communicate or not. This computational process is very time consuming, so it won’t work well in a large problem. In this paper, we first study communication based on the framework of I-DIDs, then discuss when agents suppose to communicate considering the cost and the limit resource. Experiments show that our communication algorithm works efficiently in tiger problem. We conclude that communication not only can improve the total rewards, but picking the right time to communicate is also beneficial to agents.

He Wu, Jian Luo, Le Tian

Chapter 36. A No-Reference Remote Sensing Image Quality Assessment Method Using Visual Information Fidelity Index

A novel image quality assessment method for remote sensing image is presented in the paper. Blur and noise are two common distortion factors that affect remote sensing image quality. Those two factors influence each other in both space and frequency domain. So it is difficult to objectively evaluate remote sensing image quality while exist these two kinds of distortion simultaneously. In the proposed method, the input image is first re-blurred by Gaussian blur kernels and also re-noised by white Gaussian noise. Then we measure the amount of mutual information loss before and after image filtering and noising. We take the VIF index as a measure of the information loss. The proposed method does not require reference image and can estimate distorted image with both blur and noise. Experimental results of the proposed method compared with other full-reference methods are presented. It is an accurate and reliable no-reference remote sensing image quality assessment method.

Yu Shao, Fuchun Sun, Hongbo Li

Chapter 37. 3D Model Feature Extraction Method Based on the Partial Physical Descriptor

With the rapid development of 3D scanners, graphic accelerated hardware and modeling tools, the application of 3D model databases is growing in both numbers and size. There is a pressing need for effective content-based 3D model retrieval methods. In this paper, a novel 3D model retrieval system called Physical Descriptor (PDD) is proposed. The physical descriptor is defined as the physical features extracted from the 3D surface. Firstly, after pose normalization for 3D database, the 3D model is partitioned into several parts by the planes paralleling the XOY, YOZ and XOZ plane respectively. Each partial part is represented by a physical feature named as PPD, which is a combination the inertia moment, elastic potential energy and the density of the sliced part. Several retrieval performance measures demonstrate that the proposed approach is superior to other methods.

Kuansheng Zou, Haikuan Liu, Zengqiang Chen, Jianhua Zhang

Chapter 38. Leaf Classification Methods Based on SVM and SIFT

In this research, Support Vector Machine (SVM) and Scale-Invariant Feature Transform (SIFT) are used to identify plants. For each leaf image, the algorithm localizes the keypoints and assigns orientations for each keypoint. Then it matches the sample leaves with the comparison leaves to find out whether they belong to the same category. After conducting edge detection and feature extraction, the experimental result shows that the method for classification gives average accuracy of approximately 99 % when it is tested on 12 descriptive features.

Yida Ye

Chapter 39. Saliency Preserved Image Fusion Using Nonsubsampled Contourlet Transform

The visual attention model inspired by the early primate visual system is a very important tool in image processing. Based on the visual attention model, the paper proposes a novel saliency preserved image fusion algorithm with a nonsubsampled contourlet transform (NSCT). The basic idea is that the visual saliency map is first built on the coefficients of the NSCT using the visual attention model, and then is combined with the coefficients of the NSCT to form the activity level which is employed to select the final fused coefficients. The algorithm can transform successfully the visual sensitive information from source images into the fused image which contains abundant detailed contents and preserves effectively the saliency structure while enhances the image contrast. Experiments demonstrate that the proposed algorithm yields the encouraging results.

Liang Xu, Junping Du, Qingping Li, JangMyung Lee

Chapter 40. Adaptive Wavelet Packet Filter-Bank Based Acoustic Feature for Speech Emotion Recognition

In this paper, a wavelet packet based adaptive filter-bank construction method is proposed, with additive Fisher ratio used as wavelet packet tree pruning criterion. A novel acoustic feature named discriminative band wavelet packet power coefficients (db-WPPC) is proposed and on this basis, a speech emotion recognition system is constructed. Experimental results show that the proposed feature improves emotion recognition performance over the conventional MFCC feature.

Yue Li, Guobao Zhang, Yongming Huang

Chapter 41. A Novel Decision-Based Algorithm for Removal of Highly Corrupted Images

The major drawback of recent image filtering algorithms is lack of the ability of removing high density salt-and-pepper noise. To alleviate this limitation, an improved decision-based algorithm is proposed. Firstly, according to the characteristics of salt-and-pepper noise and local gray-scale feature of pixels, this algorithm separates noise pixels and signal pixels. Then the noise pixels are recovered by the median value of the neighboring noise-free pixel values, while the signal pixels hold their gray values without changing. Different gray-scale and color images have been tested by using the proposed algorithm (PA), simulation results show that this method has the better ability of removing noises and preserving the partial details of images in comparison with some recent methods especially when the noise density is very high.

Yiyan Wang, Zhuoer Wang, Di Zhou

Chapter 42. Research and Design of Process Data Warehouse for Business Process Assessment

Business process optimization occupies a very important position in any corporation, the root of optimization lies in the mining of process logs. However, the log data in different information systems is often heterogeneous, for ease of process mining, they must be processed together in an united way, and data warehouse (DW) technology is the best choice. Appropriate process warehouse structure becomes the key to the research. Based on existing business process assessment model, this paper designs a process assessment-oriented process warehouse for storage and management of process instance data. It can be applied to the process mining, assessment and optimization of event logs. Finally, a case study demonstrates the concept given by this paper.

Hui Xia, Qing Yao, Fei Gao

Chapter 43. A Novel Emergency Cross-Media Information Retrieval Model

Existing information retrieval approaches provide only limited capabilities to capture the query requirements. However, a complete understanding of search requirements is essential for improving the effectiveness of retrieval in the emergency management field. To achieve this goal, we proposed a novel emergency cross-media information retrieval model, which includes four parts: information collection, information indexing, information retrieval and intelligent mobile terminal. The proposed model has two advantages. One is to use ontology technique to identify appropriate semantic information according to query words. The other is to use image semantic analysis based on SIFT to achieve the task of emergency image retrieval. Conducted experiments show that our model obtained encouraging performance results.

Lingling Zi, Junping Du, Qian Wang, Jangmyung Lee

Chapter 44. A Framework for 3D Model Acquisition from Multi-View Images

3D model acquisition is a fundamental issue in computer graphics and computer vision. However, constructing 3D model manually using software such as 3D MAX and Maya is a tedious and expensive work. Therefore, finding out how to obtain 3D model directly from the real world becomes a hot research topic. In this paper, we describe a framework for obtaining 3D model from multi-view images of a real object. We start with images of an object taken from different views, and then feature points extracted and matched. From the correspondences, camera calibration data and 3D geometry are acquired. Experimental results in the end of the paper show the effectiveness of the framework.

Chunmei Duan

Chapter 45. Vehicle Tracking Based on Nonlinear Motion Model

How to detect and track vehicle in nonlinear motion is a big problem in computer graphics technology. Monocular vision is complete works like positioning with only one camera; it has a simple structure and a strong applicability. In this paper, we build and solve models of traffic video with nonlinear motion model, to locate and track vehicles in video. We present a new nonlinear movement model using shape model and projection model. The accurately calculated results of position and velocity of the vehicle is given by estimating the vehicle’s speed, posture, the last position. It’s an effective solution to the problem that nonlinear movement is difficult to estimate by KALMAN filter and the tracking process is not stable and reliable enough. The experiments show that the new algorithm has better robustness than EKF and UKF algorithm.

Fan Zhang, Hong Li, Kalilou Kone, Wei Zhang

Chapter 46. Detecting Pedestrian Using Motion Information and Part Detectors

A pedestrian detection method based on motion information and part detectors is proposed in this paper used for handling partial occlusions of pedestrian in video. Extracting motion areas in the video image by fast frame difference image Gaussian mixture model as the candidate region of the pedestrian firstly; Then the part detectors including head, head-left should, head-right should, torso and so on, which have trained by the liner SVM combined with the HOG features pyramid were used to scanning detect in each candidate region individually. Finally, the Max Margin Hough Transform is used to verify the detection result. Experiments on databases and the video shoot by us show that our method has high performance in detecting pedestrians with partial occlusion.

Lingli Xu, Zhiping Zhou

Chapter 47. Tracking Algorithm Based on Joint Features

In the study of target tracking process, when the target has a similar color to the background easily leads to the loss of the target due to illumination and noise. In order to avoid the drawback of Mean shift which only uses color information as the features to track the target, Sobel operator and local binary patterns (LBP) are combined to extract the textures of the moving target as Mean shift characteristics. An advantage of the Mean shift algorithm can compute the histogram easily. However, this process can’t change the size of a search window. Therefore, the proposed method extracts the feature points of the object in the region that given by the improved Mean shift and according to the information that the positions of the special feature points, a new search window is generated. The experiments show that the proposed object tracking system performs more accurately than the Mean shift algorithm.

Xiaofeng Shi, Zhiping Zhou

Chapter 48. Distributed Audit Secure Data Aggregation for Wireless Sensor Networks

Data aggregation can reduce the communication overhead and energy expenditure of sensor nodes, as well as extend the life-cycle of the wireless sensor network. However, because individual sensors may be compromised, the data aggregation also introduces some risks including the false data injection attacks. This paper proposes a distributed audit secure data aggregation protocol. The aggregates are audited at the next level nodes of the aggregators. The communication overload, which Base Station (BS) originates in the attest process, can be avoided. Furthermore, because we can find the false data in the lower level, it is easier to strike out the false data, and only a little fraction of readings are dropped off. To do these, the aggregators attach multi-certificates to the aggregates. Those certificates may include the maximum, minimum, mean reads and those nodes’ identifiers. To further reduce the communication overload, we use the watermark method to embed the multi-certificates in authentication part of aggregates. The length of message is kept as same as that under the normal hop-by-hop aggregation protocol with MACs. The analysis shows that our protocol is efficient and provides certain assurance on the trustworthiness of the aggregation result.

Zhengdao Zhang, Zhiping Zhou

Chapter 49. Multi-Expression Based Gene Expression Programming

Among the variants of GP, GEP stands out for its simplicity of encoding method and MEP catches our attention for its multi-expression capability. In this paper, a novel GP variant-MGEP (Multi-expression based Gene Expression Programming) is proposed to combine these two approaches. The new method preserves the GEP structure, however unlike the traditional GEP, its genes, like those of MEP, can be disassembled into many expressions. Therefore in MGEP, the traditional GEP gene can contain multiple solutions for a problem. The experimental result shows the MGEP is more effective than the traditional GEP and MEP in solving problems.

Wei Deng, Pei He, Zhi Huang

Chapter 50. The Signal Processing Method of Mixed Interference Distributed Fiber-Optic Long-Distance Pipeline Leaks Detection System

A new measuring structure based on the principle of Sagnac and Mach–Zehnder mixed distributed fiber-optic interferometer can detect pipeline leakage and locate leakage point in real-time. But the actual pipeline leakage signal is non-linear and tiny broadband signal, and there is a lot of background noise in the testing environment. The null spectrum extracted to decide the location of the leak is relatively difficult. In this paper, it’s based on the discrete wavelet and least squares curve fitting method for leakage signal preprocessing. Reusing the signal through multi-scale decomposition of each signal multiplication method to determine leakage null spectrum, and the good result is achieved.

Zhengsong Hu, Qihua Yang, Qiang Wang, Renjie Zhang

Chapter 51. Multiple Faces Tracking via Statistical Appearance Model

Recently, appearance based methods have become a dominating trend in tracking. For example, tracking-by-detection models a target with an appearance classifier that separates it from the surrounding background. Recent advances in multi-target tracking suggest learning an adaptive appearance affinity measurement for target association. In this paper, statistical appearance model (SAM), which characterizes facial appearance by its statistics, is developed as a novel multiple faces tracking method. A major advantage of SAM is that the statistics is a target-specific and scene-independent representation, which helps for further video annotation and behavior analysis. By sharing the statistical appearance models between different videos, we are able to improve tracking stability on quality-degraded videos.

Jie Hou, Yaobin Mao, Jinsheng Sun

Chapter 52. A Fast Dictionary Training Algorithm for Single Image Super-Resolution

Generally the dictionary for single image super-resolution is trained by iterations of MP algorithm and K-SVD algorithm. Using the dictionary, low resolution images can be restored to high resolution images with high quality. But the training process always takes a lot of time. So in this paper we use SVD to analyze the space relationship between the high and low resolution samples, and present a cluster based algorithm for dictionary training. Compared with the K-SVD based algorithm, the proposed algorithm trains the dictionary with a much higher speed, and restores the images with similar visual quality.

Yiliang Lv, Jiwei Liu

Chapter 53. Inner-Knuckle-Print Verification Based on Guided Image Filtering

This paper presents a new approach for inner-knuckle-print verification. Firstly, guided image filtering is implemented to remove noise and the minute lines. Then robust line features are extracted from the image based on a derivative edge detector. Finally the binary line images are matched by using a cross-correlation-based method. The experiments on a finger image database which includes 2000 images from 100 different individuals show good performance of the proposed approach.

Ming Liu, Jun Yan

Chapter 54. Face Recognition Using Sequence Combinatorics

This paper presents a sequence similarity, called all common subsequences (ACS), for use with support vector machine (SVM) and k-nearest neighbors (kNN) to the face recognition problem. We first decompose face images as row and column sequences. Then use ACS, which compares two sequences by counting the number of occurrence of common subsequences, to measure the similarity of each pair of corresponding sequences in two images and the average of similarity of all pairs of sequences is proposed to be the similarity of two images. Experiments on four public face databases: Caltech, Jaffe, Orl and Yale databases, demonstrate that ACS can achieve higher recognition accuracy than some classic face recognition methods, e.g. 2DPCA and 2DLDA.

Chunhui Wang, Ankang Hu, Fenglei Han

Chapter 55. Distinction of Breast Tissues Based on Segmented Integral Area of Frequency-Resistance Curves

Breast cancer seriously endangers the health of women, which makes intra-operative assessment of cancer focus have vital significance. The information of bioelectrical impedance has unique ability to distinguish cancerous and normal tissue, and can provide basis for intra-operative assessment of cancer focus. In order to achieve accurate measurement, a measurement system is established composed of the impedance analyzer and probe with optimized electrode. Segmented integral area is regarded as characteristic parameter to reflect the over all trend. To utilize the advantages of different frequency-resistance curves, BP neural network is finally selected and good-training neural networks are integrated to make the final decision. The result indicates that the characteristic parameter selected can reflect differences of tissues and the integrated BP neural network has better performance than single neural network.

Chao Wang, Yiming Wei, Ruifeng Bai

Chapter 56. Vehicle Discrimination Using a Combined Multiple Features Based on Vehicle Face

In this paper, a new method for vehicle discrimination and recognition on the basis of combination of multiple features based vehicle face is proposed. The color difference, vehicle face pattern difference and logo matching degree are getting together to improve the performance of vehicle discrimination. This method is assessed on a set of 200 images that belong to five distinctive vehicle manufacturers. A series of experiments are conducted, splitting the 300 pairs of images to a training set and a test set, respectively. It is shown that the enhanced feature combination approach (CMN) proposed in this paper boosts the recognition accuracy compared with the CM and CN method. The reported results indicate a high classification rate in similar or different vehicles and a fast processing time, making it suitable for real-time applications.

Yingnan Wang, Hong Li, Clement Kipkorir Kirui, Wei Zhang

Chapter 57. Fast SIFT Algorithm Using Recursive Gaussian Filters

Scale invariant feature transform (SIFT) algorithm has drawn great attention from computer vision engineers since it was proposed in 1999. However, the high computational complexity of the algorithm has hindered its application. In this paper, a fast SIFT algorithm is proposed, in which FIR Gaussian filters are replaced by recursive filters. Experimental results show that the proposed fast SIFT method needs less computation and yields nearly the same performance compares to original method. It is also recognized that the impulse response approximation error can be used as a good measure to estimate performance degradation of SIFT algorithm in recursive Gaussian filters. Furthermore, through using recursive filters, more choices of the values of prior smoothing scale can be made without considering the number of operations.

Zhengyuan Ye, Shouxun Liu, Xuan Wang

Chapter 58. A Robust Linear Camera Calibration Based on Coplanar Circles

In this paper, a novel method for camera calibration based on coplanar circles is proposed. Firstly, we describe geometrical interpretation of the proposed method. Then the corresponding strict algebraic proof is presented. Experiment shows that the result is accurate and the proposed strategy is robust.

Yu Cai, Yanjin Huang

Chapter 59. A Multi-Classification Algorithm of Semi-Supervised Support Vector Data Description Based on Pairwise Constraints

A constraint-based semi-supervised support vector machine classification learning algorithm is proposed based on support vector data description algorithm with pairs of semi-supervised learning thinking combined. Multiple hyperspheres are constructed by constraints for the


-classification problems, so that the original problem converted to a


-classification problem. The algorithm to get positive constraints label and negative constraints label by calculating the degree -membership of unlabeled samples, then multiple hyperspheres constructed based on the multi-classification algorithm. Finally, simulation experiments on artificial datasets and UCI datasets to verify the effectiveness of the algorithm.

Ying Zhao, Guan-jun Wang

Chapter 60. Real-Time Vehicle Classification Based on Frequency Domain Energy Spectrum

Vehicle classification is now an important part of Intelligent Transportation Systems (ITS). Especially in toll station and parking, real-time vehicle classification technology is used to determine the vehicle information. A novel method based on frequency domain energy spectrum of geomagnetic sensor for real-time vehicle classification was proposed in this paper. According to the definitions of eight frequency domain energy formulations, the energy values with different frequency regions could by computed. Compared with those energy values, the optimal frequency region and energy formulation were obtained. As each vehicle classification has a specific energy region, the classification of each vehicle can be easily differentiated by its energy value. Results show that the vehicle classification method proposed in this paper has an excellent performance and the average accuracy is more than 90 %. Besides, the algorithm makes it easier for applications in sensor nodes with limited computational capability and energy source.

Pengfei Zhang, Haijian Li, Honghui Dong, Limin Jia, Maojing Jin

Chapter 61. A New Combination Sampling Method for Imbalanced Data

Imbalanced data is commonly in the real world and brings a lot of challenges. In this paper, we propose a combination sampling method which resamples both minority class and majority class. Improved SMOTE (ISMOTE) is used to do over-sampling on minority class, while distance-based under-sampling (DUS) method is used to do under-sampling on majority class. We adjust the sampling times to search for the optimal results while maintain the dataset size unchanged. Experiments on UCI datasets show that the proposed method performs better than using single over-sampling or under-sampling method.

Hu Li, Peng Zou, Xiang Wang, Rongze Xia

Chapter 62. Breast Tissue Segmentation Using KFCM Algorithm on MR images

Breast MRI segmentation is useful for assisting the clinician to detect suspicious regions. In this paper, an effective approach is proposed for segmenting the breast into different regions, each corresponding to a different tissue. The segmentation work flow comprises three key steps: MR Images preprocessing, locating breast-skin and breast-chest wall boundary by using OTSU thresholding algorithm, and segmenting fibroglandular and fatty tissues with applying the kernel-based fuzzy clustering algorithm (KFCM). The proposed method was applied to segment the clinical breast MR images. Experimental results have been shown visually and achieve reasonable consistency.

Hong Song, Feifei Sun, Xiangfei Cui, Xiangbin Zhu, Qingjie Zhao

Chapter 63. Information-Theoretic Clustering for Gaussian Mixture Model via Divergence Factorization

Multivariate Gaussian distribution is a multi-dimensional generalized model for conventional pervasive Gaussian distribution and Gaussian mixture model (GMM) is a practical model widely used in clustering analysis. In this paper, we aim at clustering the Gaussian mixture components into reduced number of components, thus simplifying the GMM. By utilizing a class of Bregman divergences in information-theoretic context, we minimize the in-cluster relative entropy, and derived a closed-form of optimal parameters by divergence factorization. Symmetric relative entropy is used to avoid the asymmetric manner of divergence. Our algorithm is time-saving as a simplification of GMM and demonstrates its superiority visually in practical clustering-based image segmentation problem, in comparison with conventional methods, such as k-means and GMM-EM.

Jiuding Duan, Yan Wang

Chapter 64. A Method of Acceleration Applied in Symmetric-SIFT and SIFT

Symmetric-SIFT is an effective technique used for registering multimodal images. It is based on a well-known image registration technique named Scale Invariant Feature Transform (SIFT). Similar to SIFT, Symmetric-SIFT detects many stable keypoints even though not all of which are useful. Experiments show that matching keypoints are mostly on or near the edge. Based on the phenomenon, we propose an effective method. In our method, We extract the edge and classify keypoints by whether they are on the edge or not. Then we delete the points that are far from the edge and match the remained ones. Finally, we get the matching set after filtering the initial matching result with the threshold value got by the Bayesian formula. The experimental results show that the proposed method can not only greatly reduce the matching time, but also effectively improve the matching rate.

Dong Zhao, Qi Wang, Haiyan Sun, Xiaopeng Hu

Chapter 65. A Novel Method of Image Enhancement via Multi-Scale Fuzzy Membership

For the imaging condition restriction, nature images sometimes have the problem with low contrast and low illumination. In order to solve those problems, we proposed a novel image enhancement algorithm based on multi-scale fuzzy membership. The images will be firstly decomposed into multi-scale sub images by Laplacian pyramid and then be enhanced through calculating the fuzzy membership degree under multi-scale. Finally the edge-preserving image denoising will be conducted by bilateral filter to realize the effectively enhancement of the nature images. The experimental result shows that the proposed algorithm also can be used in X-ray medical image enhancement and achieves a better effect compared with the traditional method and has certain theoretical and practical application value.

Ce Li, Yannan Zhou, Chengsu Ouyang

Chapter 66. Adaptive Region Clustering in LDA Framework for Image Segmentation

Image segmentation based on low-level features has been studied for many years. However, because of the semantic gap issue, it is difficult to have more breakthroughs based on low-level features. LDA is a powerful tool to model co-occurrence relationships between words and thus is used to catch the semantic connections between low-level visual features. In image segmentation, the codebook is built from the visual features and topics are trained with LDA model. And the topic distributions yield important guidance for segmentation. However, in previous papers, researchers used the topic with the highest probability to merge the regions. It ignored the statistics nature of the topic distribution. And, the segmentation result will be greatly impact by the codebook size, the topic number and cluster number. To address these challenges, this paper proposes a new image segmentation algorithm based on LDA framework: an adaptive region clustering approach based on EM. We build the cookbook from the color, texture and SIFT features and perform the LDA training using Gibbs Sampling for topics. Then the adaptive region clustering with EM is invented to merge the regions based on topic distribution. The clustering number is self-identified according to Minimum Description Length (MDL) principle. And an image is represented as a Gaussian Mixture Model (GMM) with objects corresponding to Gaussian mixture components. The final segmentation could be achieved after the region clustering and adjacent check. We implemented the new algorithm and conducted experiments to validate the region clustering approach and segmentation performance. And the results show great effectiveness of this new algorithm.

Xiaoru Wang, Junping Du, Shuzhe Wu, Fu Li

Chapter 67. Methods of Recognizing True and Fake Smiles by Using AU6 and AU12 in a Holistic Way

Smile is one of the simplest forms of expressions that it is easy to recognize for human beings. It will be one of the most natural, straightforward and friendly ways in Human Computer Interaction (HRI) if a computer could catch the subtle expression, understand the inner state of human and meanwhile give its feedback according to the corresponding instance. In this paper, some different methods are proposed, to realize the recognition of true and fake smiles, based on facial action units from the research field of psychology and human behavior. In all of the methods we used, AU6 and AU12 are dealt with together in each example, which is different from AU recognition. Some popular feature extraction and classification methods such as Gabor wavelets, 2DPCA, Adaboost and SVM are used in the holistic way to implement the recognition. Images in our database are all frontal facial images with smiles of different types and levels from subjects of different countries with different colors and ages. Lots of experiments show that the best accuracy of our methods in recognizing true and fake smiles is close to 86 %, while people’s true-fake-smile recognition ability is much lower.

Pingping Wu, Wenmin Wang, Hong Liu

Chapter 68. Constrained Silhouette Based Evolutionary K-Means

Evolutionary K-Means (EKM) is a non-parametric approach proposed to improve K-Means algorithm. Current EKM approaches are ineffective in deciding the correct cluster number of real datasets. This paper uses instance-level constraints to solve this problem and presents a Constrained Silhouette (CS) based algorithm, namely CS-EAC. Firstly CS is defined to combine constraints into the computation of Silhouette Information (SI). Updated from the Fast Evolutionary Algorithm for Clustering algorithm (F-EAC), CS-EAC uses CS instead of SI to guide the genetic operations. Experimental results suggest that CS-EAC is effective in both deciding the correct number of clusters and improving the accuracy of clustering for real datasets.

Zhenfeng He

Chapter 69. Traffic Light Detection and Tracking Based on Euclidean Distance Transform and Local Contour Pattern

This paper proposes a new recognition approach for traffic light based on Euclidean distance transform (EDT) and local contour pattern (LCP). There are two main contributions of this paper. First, this paper combines principle component analysis (PCA) with EDT-based image to detect traffic light colors. The color space for specific colors is partitioned more precisely, which leads to a high recognition rate. Second, we incorporate the above color detection into the contour segmentation of traffic light holder based on the LCP to further improve the recognition rate of traffic light. The experimental results show that our approach is able to detect traffic light far away from camera about 50–80 m and the average recognition rate can reach up to 99.29 %.

Zhenyang Wang, Zhidong Deng, Zhen Huang

Chapter 70. Multi-Correlation-Based Mode Decision for Multi-View Video Coding

In this paper, an efficient mode decision algorithm, named the multi-correlation-based mode decision (MMD), is proposed to reduce the computational complexity while maintaining the high coding efficiency. In this method, the rate distortion (RD) cost of the Direct mode is always computed and compared with an adaptive threshold as a possible early termination chance. This adaptive threshold is determined by using the spatial, temporal and inter-view correlation between the current macroblock (MB) and its neighboring macroblocks. Experimental results demonstrate that the proposed MMD algorithm can significantly achieve computational saving of 72.38 % on average with no significant loss of rate-distortion performance, compared with the full mode decision in the reference software of multi-view video coding (MVC).

Qinghong Shen, Fengsui Wang, Sidan Du

Chapter 71. Automated Discrimination of Gait Patterns Based on sEMG Recognition

A general scheme of automated discrimination of gait patterns based on recognition of surface electromyogram of lower limbs is proposed to classify three different terrains and six different movement patterns. To verify the effectiveness of different feature extraction methods, time–frequency features such as RMS and MF, wavelet variance and matrix singularity value are employed to process the sEMG signals under different conditions. SVM is used to discriminate gait patterns based on the selected features. Comparison results indicate that feature extraction method based on matrix singularity value can obtain better results and over 92.5 % classification accuracy ratio can be achieved. Experimental result indicates the rationality and effectiveness of the proposed methods for feature extraction and pattern classification. The proposed scheme shows great potential in the application of lower limb assistance.

Fei Wang, Xiao Hao, Baoxiang Zeng, Chucheng Zhou, Song Wang

Chapter 72. Hierarchical Sparse Representation for Traffic Sign Recognition

Researchers have proposed various machine learning algorithms for traffic sign recognition (TSR), which is a supervised multicategory classification problem with unbalanced class frequencies and various appearances. This paper presents a novel framework for traffic sign recognition exploiting the sparse property of intrinsic information of traffic sign. The contributions of our work are twofold: on one hand, the intrinsic discriminating information among different categories is utilized, on the other an efficient hierarchical sparse representation classification (HSRC) strategy is adopted. Experiments on publicly available datasets show that HSRC is efficient for traffic sign recognition, achieving higher accuracy than many state-of-the art schemes.

Yaxiang Fan, Hao Sun, Shilin Zhou, Huanxin Zou

Chapter 73. A New Sequence-Based Approach for XML Data Query

In order to avoid expensive join operations in query processing from structured XML document, some index methods based on sequence have been proposed, which transform XML documents and twig patterns into sequences. By performing subsequence matching, query is processed holistically without breaking the twig pattern into many individual root-to-leaf paths, and large useless intermediate results and expensive join operations are avoided. In this paper, combining path sequence strategy with region labeling scheme, we propose a new sequence scheme, Region Path sequence scheme, where the last node of each path is labeled with the region labeling scheme. Compared with previous approaches, our approach can avoid false alarm more effectively, and any extra structure for labeling needn’t be constructed. Furthermore, we construct two level B


-tree structure to finish the matching, and also propose corresponding matching algorithm. Experiment results demonstrate that our approach can not only avoid false alarm, but also process query more quickly than previous methods such as ViST, and Constraint Sequence.

Wen Li, Jin Yang, Gaofeng Sun, Sen Yue

Chapter 74. Model-Based Workpiece Positioning for Robotic Fixtureless Assembly Using Parallel Monocular Vision System

This paper proposed to use parallel monocular vision system that could fit different robotic grasping pattern, in order to reduce the computation burden for real-time grasping control. A novel model-based workpiece positioning approach, which can solve both 3D or 2D pose estimation problem, is proposed by using the imagery template and homography matrix. The demand of workpiece template is not 3D model, but the workpiece template image, which is much easier to obtain. Moreover, as the positioning expressions based on homography matrix between the workpiece template and images, and the two camera images, are expressed as a simple formula, the proposed approach is intuitive for algorithm development.

Weiwei Yu, Mingmin Zhai, Yasheng Chen

Chapter 75. Single-Trial Identification of Motor Imagery EEG based on HHT and SVM

Single-trial identification of motor imagery (MI) EEG is one of the key techniques in the brain-computer interface (BCI). To improve the accuracy of classification and reduce the algorithm time, targeting at motor imagery (MI) EEG of four kinds of motion, a single-trial identification algorithm of MI EEG based on HHT and SVM is proposed. Firstly, MI EEG is decomposed into 8-order intrinsic mode function (IMF) and margin R by empirical mode decomposition (EMD). Secondly, Hilbert spectrum is got by Hilbert transformation. AR model parameter of the extracted 6-order IMF is extracted. The acquired 6-order AR parameter and the characteristic quantity of 29 power spectral density included in the 4-32 Hz EEGs constitute a 35 dimensional characteristic vector. Finally, support vector machine (SVM) is used to classify. The single-trial identification results are as follows: the average recognition rate of the two kinds of thinking actions is 91.6478 %, and that of three is 89.4798 %, four is 89.4064 %.

Peng Lu, Daoren Yuan, Yafei Lou, Chi Liu, Shilei Huang

Chapter 76. Robust Visual Tracking Using Incremental Sparse Representation

The sparse representation has achieved considerable success in visual tracking due to its simplicity and robustness. It requires each target candidate is sparsely represented in the space spanned by target templates and trivial templates. The sparsity is achieved by solving an l1-regularized least squares problem. When the sparse representation is incorporated into the framework of particle filter, solving l1 minimization problem for each particle independently requires a large calculation time, making real-time implementation difficult. In this paper, we exploit the redundancy between particles and use the homotopy method to design an incremental likelihood function calculation approach, and therefore form an efficient and robust visual tracking algorithm. The proposed algorithm is tested on extensive video sequences and the experimental results are found to be highly competitive with other recent trackers.

Song Pan, Huaping Liu

Chapter 77. A Study on Design of Early Warning Decision Support System of Desertification

At present, the study on the slowly varying monitoring and early warning system is not perfect at home and abroad. In this paper, desertification database, knowledge base and model base are designed in details and the future evolution of the desert is predicted by the desertification early warning system. The results show that correct rate of simulated distribution of desertification reaches over 90 % in the study area. And ‘Auto Set Sand instrument—Monitoring Platform’ can conduct sand data analysis at anytime. In addition, the WebGIS based system provides necessary decision support for the government.

Zhengwei Li, Jing Du, Xianyong Meng, Chen Sun, Yongqiang Liu

Chapter 78. Fast Fusion Method of TT&C Data with Multi-Routing Transmission Model

In view of the problem of valid data identification brought by the multi-routing transmission model in the new generation of TT&C IP network, the characteristic of TT&C IP network is analyzed. According to the requirements of transmitting TT&C data in real time and with high reliability, the short delay priority principle is taken, the comparison operator of TT&C data is defined, and the algorithm of fast fusion of multi-routing data suitable for the receiver node is given and tested. The results show that the algorithm has strong points of short delay and high real-time performance.

Bin Tian, Yue Yang, Yanhui Pan, Shengjun Luo

Chapter 79. Discovery of Static Test Configuration Model and Data Model Based on TTCN-3 Test Systems

Aimed at the comprehensibility, reusability and maintainability, the thesis presents the reverse model recovery for the legacy code developed by TTCN-3. It can also help tester and maintainers to verify the test implement, etc. The thesis introduces the discovery of static test configuration model and data model based on the reverse model discovery system framework.

Yongpo Liu, Shuangmei Liu, Ji Wu, Chuangye Chang

Chapter 80. A New Method of Unknown Radar Signals Sorting

Aimed at the problem of worse real-time in the sorting of unknown radar signals, a new signal sorting method based on the DBSCAN is put forward. Using the parameter distribution characteristics of radar pulse data, the method firstly searches a number of reference points to properly represent original data points, and then, make use of their density connected character to cluster the reference points. Because of the decrease of the data involved in clustering, the method overcomes the immense calculated quantities of original algorithm, and enhances the clustering speed of unknown radar signal sorting. Computer simulation results show that the proposed algorithm can effectively sort the unknown radar signals.

Xiaofeng Wang, Xuzhou Zhang, Runlan Tian, Xinglong Qi

Chapter 81. The Creation of Interactive Three-Dimensional Microscope Based on VRML and JavaScript

When constructing sophisticated 3D interactive apparatus with VRML, a method is usually adopted that is manipulating another object with a selected one. Therefore, the way of achieving the linkage movement of three-dimensional objects on website was crucial. Adjustment of the apparatus for complex interaction can not simply be operated with VRML itself. This article takes the creation of interactive three-dimensional microscope for example, with the technique of JavaScript programming which realized the interactive operation adjusting the place of microscope lens horizontally and vertically, a complex interaction and coordinated controlling campaign achieved.

Di Wu

Chapter 82. Research and Implementation of Face Detection System on Android Smart Phone

In order to improve the effect of detecting faces in Android smart phone, we proposed an effective face detection algorithm. Cluster analysis is used to segment skin region from the input color image in YCbCr color space. After the normalization of candidate regions being finished, the weighted Euclid distance is calculated between the reference templates and candidate regions. Based on this algorithm we implemented a face detection system on Android platform. Experiment proves that the system allows robust and has a high detection rate.

Xin Li, Yumei Zhai, Xiong Li

Chapter 83. Image Mosaic Based on SURF and Results Optimization

In order to improve the insufficiency of SURF algorithm, we present an auto-adjusted algorithm of image size based on phase-correlation. We detect the zoom relationship and translation co-efficiency between the images and modulate the unregistrated image’s scale to the same level as the original image. We obtain the Region of Interest (ROI) according to the translation parameter and then pre-treat the images. We propose an image matching method based on the saliency map. We calculate the saliency map in the Region of Interest and mark the interest points in the area by using SURF algorithm. In the part of image fusion, the method of gradated in-and-out which we commonly use is improved to eliminate seam. As the image exposure differences, we propose two methods to adjust the exposure of spliced image. One is based on mean value and the other is based on mean–variance specification. Through the experiment, the validity of the proposed methods is demonstrated and the quality of the mosaic image is better.

Tie Jiang, Guibin Zhu

Chapter 84. Improved Blind Source Separation Based on Non-Holonomic Natural Gradient Algorithm with Variable Step Size

The traditional natural gradient algorithm works badly when the source signal amplitude changes rapidly or becomes zero at a certain time. In addition, it cannot resolve very well the contradiction between the convergence speed and the error in steady state because the step-size is fixed. In order to solve the above problems, this paper proposes an improved blind source separation algorithm based on non-holonomic natural gradient by choosing an adaptive step-size and a suitable nonlinear activation function. Simulation result demonstrates that the new algorithm performance is superior to the traditional natural gradient algorithm.

Ce Ji, Baocheng Tang, Kun Yang, Mingbo Sha

Chapter 85. Multi-Level Fingerprint Classification Based on Average Frequency of Ridges for Large Scale Fingerprint Database

In order to improve the recognition speed, accuracy and robustness of the Automatic Fingerprint Identification System based on large scale fingerprint database, a multi-level fingerprint classification method was proposed based on three independent classification features, in which the quality evaluation indexes algorithm of fingerprint images was introduced to evaluate the quality of input fingerprints. For those good quality fingerprints, we classify them into three features according to fingerprint pattern type, ridge count between singular points and average frequency of ridges in central region, respectively, which could decrease the retrieval space gradually. Experimental results on NIST DB4 show that the proposed classification algorithms with redundancy mechanism have high retrieval efficiency and strong robustness, which provides a rapid and effective index approach for the large scale fingerprint database.

Xiaoqi Peng, Yunfei Zhong

Chapter 86. Two Improved Edge Coloring Algorithms for Data Migration

The problem of computing a migration plan among the store devices for moving the data from current configuration to the target configuration is called data migration problem. There are some reasons to cause the data migration, the reordering of the data combination, the system’s load balancing, and the change of use mode. In this paper, we ignore the network speed and transfer speed. And only consider the situation that each store device can be used as sender of receiver in the transfer process at the same time. We develop two algorithms based on the multi-graph edge coloring problem: complete decomposition algorithm and the Greedy of maximum degree and weigh match algorithm. They can be used in different circumstances and has better performance, they have better parallelism in the data migration, and consume less time in the process of data migration.

Gangfeng Huang, Maishun Yang, Mingming Jing

Chapter 87. Medical Image Fusion Algorithm Based on the Laplace-PCA

Medical image fusion processing, as an indispensable part of the modern medical treatment, has been used widely in clinic medicine. The Paper firstly describes the Gauss decomposition of the medical image and the establishment of the Laplace pyramid decomposition images, then fusing medical image using PCA fusion criteria, finally obtaining the fusion image based on the Laplace pyramid image reconstruction. The experimental results show that the algorithm can be complementary information of the CT image and MR image highlights, and has a good fusion effect.

Pengtao Zhao, Gang Liu, Cen Hu, Huang Huang, Bing He

Chapter 88. Filter Parameter Estimation in Non-Local Means Algorithm

In this paper, improvements to the Non-local Means (NL-Means) algorithm introduced by Buades et al. are presented. The filtering parameter is unclearly defined in the original NL-Means algorithm. In order to solve this problem, we calculated filtering parameter by the relation of noise variance, and then proposed a noise variance estimate method. In this paper, noisy image is transformed by wavelet. The wavelet coefficients in each sub-band can be well modelized by a Generalized Gaussian Distribution (GGD) whose parameters can be used to estimate noise variance. The simulation results show that the noise variance estimate method is not only exact but also makes the algorithm adaptive. The adaptive NL-Means algorithm can obtain approximately optimal value, and need less computing time.

Hong-jun Li, Wei Hu, Zheng-guang Xie, Yan Yan

Chapter 89. Image Fusion Using Compressed Sensing in Nonsubsampled Contourlet Transform Domain

Image fusion algorithm using compressed sensing theory in NonSubsampled Contourlet Transform (NSCT) domain is proposed, NSCT can provide better sparsity than wavelet transform in image transform. After the transform of NSCT, low-frequency coefficients of the image are preserved, only high-frequency coefficients are measured. Fused coefficients are calculated according to different fusion rules in low frequency and high frequency domain. In the reconstruction, OMP algorithm is used to recover the high-frequency coefficients and the image is reconstructed by inverse nonsubsampled contourlet transform. Compared with wavelet compressed sensing algorithms, simulation results demonstrate that the quality of reconstructed image can be greatly improved.

Fu Liu

Chapter 90. The Reverse Loop Subdivision Algorithm on Approximate Minimum Error

A new reverse Loop subdivision algorithm based on approximate minimum error is presented. At first, all isolated reverse equations in a finite grid region are found out according vertices relationship and symmetry of Loop subdivision mesh. And an integrated reverse equation is constructed using all these isolated equations by undecided parameter method. Then, the error between original grid and re-construction grid is calculated by vary original vertex. And undecided parameters are solved by using minimization error condition. To make undecided parameters independent on valence of neighbors, some approximation is applied on calculating mean error. At last, the integrated equation’s computing effect on whole grid is analyzed and an equivalent computation is obtained, and a fast implementing steps is given. The experimental results show that our algorithm can get a stable reverse result for those distorted subdivision grid.

Boning Ma, Longxing Kong, Xiaoan Tang, Gangyao Kuang

Chapter 91. A Hybrid Algorithm Based on PBIL Algorithm and Zooming Algorithm and Its Convergence Proof

A hybrid algorithm (HA) based on population based incremental learning (PBIL) algorithm and zooming algorithm (ZA) is proposed, and its convergence is proved in this paper. In the hybrid algorithm, PBIL algorithm is employed for the evolutionary process as it can accelerate the convergence speed by a reduced time complexity, zooming algorithm is used to improve the PBIL algorithm as it can reduce search space on a large scale, and develop the convergence speed and the precision of solution obviously. The convergent analysis shows that if the population is big, and the parameters are proper, the hybrid algorithm converges to the global optimal solution.

Gaopeng Wang

Chapter 92. Image Intensity Correction Under Illumination Changes Based on Mixture Gaussian Model

Illumination changes in video sequences result in a drastic increase in the number of falsely detected change regions and make change detection unreliable. In this paper, we propose a novel approach for intensity correction under illumination variation. A mixture Gaussian model consisting of two density components associating with two classes is used. Based on Expectation–maximization algorithm, the statistical parameter estimations are performed. Under the assumption of Gaussian distribution for stationary pixels, the global intensity factor can be calculated for image intensity correction. Finally, two experiments are carried out to verify the proposed method.

Yanxiang Han, Zhisheng Zhang, Lei Zhang, Ping Chen, Fei Hao

Chapter 93. Maneuver Target Detection Method with Iterative Endpoint Fitting Assisted

A target maneuvering detection method with iterative endpoint fitting assisted is derived and presented. With the similarity between image curve fitting and maneuvering target tracking has been thought of in the method, when the traditional target maneuvering detection method was used, the target maneuver information could be abstracted with the principle of iterative endpoint fitting assisted, and the maneuver start point would be detected more accurately by the information feed backed to the detection process. The effectiveness of the method has been verified by numerical simulation.

Zhangsong Shi, Zhonghong Wu

Chapter 94. Multi-Camera Tracking via Online Discriminative Feature and Multi-Cue MRF

Visual tracking across distributed cameras with disjoint views consists of many challenges, such as illumination changing and similar appearance of multiple persons. In this paper, we present a new solution to the problem in the formulation of Multi-Cue Markov Random Field (Multi-Cue MRF), and employ the max-product linear programming (MPLP) algorithm to find the MAP configuration of MRF. Moreover, in order to bridge the gap among different camera views, we propose a hybrid strategy which integrates spatio-temporal relationship modeling, online visual feature selection and local pair-wise code (LPWC) extraction into one framework. Finally, experimental results conducted with challenging video sequences verify the effectiveness of our method.

Jianyong Wang, Feng Chen, Jianwu Dong, Dingcheng Feng
Weitere Informationen