模式识别与人工智能
Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文
Pattern Recognition and Artificial Intelligence
22 Judgement and Disposal of Academic Misconduct Article
22 Copyright Transfer Agreement
22 Proof of Confidentiality
22 Requirements for Electronic Version
More....
22 Chinese Association of Automation
22 National ResearchCenter for Intelligent Computing System
22 Institute of Intelligent Machines,Chinese Academy of Sciences
More....
 
 
2015 Vol.28 Issue.1, Published 2015-01-31
Papers and Reports    Researches and Applications   
   
Papers and Reports
1 Research on Multi-Label Propagation Clustering Method for Microblog Hot Topic Detection
CHEN Yu-Zhong, FANG Ming-Yue, GUO Wen-Zhong
With the rapid growth of microblog data, extracting hot topics from vast amounts of microblog posts has become a research hotspot. The traditional methods for hot term extraction can hardly apply to microblog data, thus a life value calculation model based on aging theory is established to extract hot terms. Then, a hot term co-occurrence network is built based on the correlations between hot terms. Aiming at the problem that traditional clustering methods can hardly handle the hot term overlap between different topics and can not deal with vast amounts of data efficiently, a term clustering method based on multi-label propagation algorithm (TCMLPA), which has a nearly linear time complexity, is proposed to detect hot topics in hot term co-occurrence network.The experimental results show that life value calculation model can filter noise and extract hot terms effectively. Meanwhile, TCMLPA ensures the stability of clustering result and improves the accuracy and efficiency of hot topic detection.
2015 Vol. 28 (1): 1-10 [Abstract] ( 728 ) [HTML 1KB] [ PDF 568KB] ( 1605 )
11 A Multiphase Level Set Method for Fast Segmentation Based on AOS Scheme
YAN Mo, SHUI Peng-Lang
The numerical implementation of level set method based on upwind scheme needs to reinitialize the level set function during the evolution of the curve. To guarantee the stability of the algorithms, a small time-step must be selected, which slows down the running speed. Based on a level set method without reinitialization, a semi-implicit additive operator splitting (AOS) scheme is used in numerical implementation,and a unified implementation for different statistical models is provided. Based on two-phase segmentation, a new level set method for multiphase segmentation is proposed. This method uses a unique level set function many times during curve evolution to realize multiregion segmentation. Its advantages are as follows. AOS scheme is used in this method which is unconditionally stable,and it allows to use a large time-step. It provides unified numerical implementation for many kinds of statistical models. The unique level set function is used in curve evolution, which reduces the amount of evolution equations and is more flexible. The experimental results show that the proposed method is more efficient and can segment multiregion images exactly.
2015 Vol. 28 (1): 11-18 [Abstract] ( 599 ) [HTML 1KB] [ PDF 1184KB] ( 1044 )
19 Zenithal Pedestrian Detection Based on Histogram of Oriented Gradient
TANG Chun-Hui
There are extensive researches on pedestrian detection, which mostly suppose visible humans observed in flat view and are applied in video surveillance,driving assistance etc. However, sometimes pedestrian detection from another perspective should be considered in practice. In this paper, a histogram of oriented gradients (HOG) descriptor is introduced for the zenithal pedestrian head detection. The vectors extracted from training samples are trained in the support vector machine to get the classifier parameters, and then the vectors of test samples are input into the classifier to discriminate which targets are. Compared with the existing methods, the proposed descriptor highlights both the region and the contour of the object. Partitioning blocks are reformed, and the feature calculation and statistical method are changed adaptively to the task. The experimental results show that the proposed method is effective and can be applied to count pedestrians in vertical view with a faster processing speed.
2015 Vol. 28 (1): 19-26 [Abstract] ( 674 ) [HTML 1KB] [ PDF 789KB] ( 1860 )
27 Trend Prediction for Microblog Based on Classification Modeling of Heat Curves
LIU Ye-Zheng, DU Ya-Nan, JIANG Yuan-Chun, DU Fei
Timely acquiring of hot topics is of great significance for commercial innovation and business marketing. Existing methods mostly need to cope with non-structured data or repeated traversal sample set, which results in high complexity. In this paper, emphasizing the topic statistical properties, a non-parameter method based on structured data is proposed to acquire the hot topics in time. Firstly, diffusion degree and focus degree are introduced to build heat curves to characterize the topics. Then, the varied heat curves are classified to determine the common behaviors of the topics. Finally, the weighted-vote scheme is employed to predict whether a topic is trend or not. The experimental results on Sina microblog show that the proposed method has simple data structure and works well with low time complexity and simple manipulation.
2015 Vol. 28 (1): 27-34 [Abstract] ( 545 ) [HTML 1KB] [ PDF 967KB] ( 1382 )
35 A Face Detection Algorithm Based on Haar-Like T Features
WANG Qing-Wei, YING Zi-Lu
An algorithm is presented for face detection based on Haar-Like T features which are the extension of Haar-Like features. Due to the distributions of facial organs, a lot of T structure features on face models can be found. Based on the principle of Haar-Like features, 4 Haar-Like T features are presented which are similar to Haar-Like features. Haar-Like T features and Haar-Like features are all input into Adaboost algorithm to generate weak classifiers for feature selection. Finally, a strong classifier is constructed by cascading those weak classifiers for face detection. Extensive face detection experiments are conducted for the proposed algorithm. Compared with the traditional face detection classifier, such as Haar-Like classifier and LBP classifier, the superior experimental results prove the effectiveness and the superiority of the proposed algorithm.
2015 Vol. 28 (1): 35-41 [Abstract] ( 740 ) [HTML 1KB] [ PDF 660KB] ( 986 )
Researches and Applications
42 Cellular Genetic Algorithm Based on Chaotic Map
LI Xue-Yan, LI Xue-Mei, LI Xue-Wei, Wu Jin-Pei
According to the function and structure characteristics of cellular genetic algorithm (CGA), chaos cellular genetic algorithm (CCGA) based on Cat map, Logistic map and Tent map are designed respectively with the organic combination of cellular genetic algorithm and chaos algorithm. Besides, the ergodicity of three chaotic mappings are explained. Taking advantage of chaotic ergodicity and sensitivity to initial condition, the initial distribution of population is optimized, the searching scope of the algorithm is enlarged, the mechanism of local chaotic crossover operator and chaotic mutation disturbance are designed, and the changes of population diversity are compared under different mapping operators. Theoretical analysis and simulation results show that the proposed algorithm has obtained good performance in improving optimizing accuracy, accelerating convergence and avoiding the local optimum by introducing three chaotic maps.
2015 Vol. 28 (1): 42-49 [Abstract] ( 485 ) [HTML 1KB] [ PDF 1140KB] ( 697 )
50 Blind Image Denoising Based on Noise Level Estimation
FANG Shuai, XIA Xiu-Shan, CAO Yang, YU Lei
Block-matching and 3D filtering (BM3D) algorithm is one of the best image denoising algorithms. However, the application of the algorithm is constrained owing to high time complexity and the requirement of exact image noise level parameter. Thus, a fast block-matching and 3D filtering (FBM3D) algorithm is proposed, which uses a grid-based block-matching strategy. Then, the image noise is refined by iteration in which the starting point is set by SVM learning and the ending point is decided by image quality. The experimental results show that the proposed algorithm has a significant improvement in computation efficiency, visual effects and quantifiable performance evaluation.
2015 Vol. 28 (1): 50-58 [Abstract] ( 892 ) [HTML 1KB] [ PDF 2137KB] ( 1562 )
59 Twice Regression Learning and Its Application on Software Effort Estimation
YANG Zi-Xu, LI Ming
Regression learning belongs to supervised learning, which is to build models on examples with real-valued labels. It usually needs a great amount of training samples to obtain significant performance. However, there are few training samples that can be collected in real applications. Aiming at this problem, the neural network ensemble to regression tree(NERT) algorithm is proposed based on the twice learning framework. By means of the virtual sample generation technique, this method makes effective utilization of two sequential learning stages to relieve the problem of insufficient training samples for enhancing its performance. By choosing two methods with high generalization ability and significant comprehensibility respectively for the two stages, a model with two characteristics can be obtained. Results on software effort estimation with few training samples show that NERT is capable of achieving better performance from these small data than existing methods, and reveals the key factors within effort estimation effectively due to its inherent comprehensibility.
2015 Vol. 28 (1): 59-64 [Abstract] ( 494 ) [HTML 1KB] [ PDF 452KB] ( 771 )
65 A Dynamic Background Modeling Based on Weighted Histogram
CHU Jun, YANG Fan, WANG Lu, ZHU Tao
The illumination variation, waving trees, rippling water and noise are the main problems for the establishing of background model of dynamic scene. Aiming at the problems, a dynamic background modeling method is proposed based on the weighted histogram. In the proposed method, a weighted histogram is firstly defined by fusing the local spatial correlation of the image sequence, and it is regarded as a feature to represent the dynamic scene. Then, a simple clustering criterion for weighted histogram is proposed, which is used to cluster features by calculating luminance and chrominance components of the weighted histogram separately. Compared with the MOG(Mixture Of Gaussians), SCBM(Standard Codebook Model), HSVCBM(HSV CodeBook Model)and WHM(Weighted Histogram Model), the experimental results on several standard test video sequences show that the proposed method has better adaptability to the dynamic scene.
2015 Vol. 28 (1): 65-73 [Abstract] ( 527 ) [HTML 1KB] [ PDF 940KB] ( 620 )
74 Fast Spatial-Temporal Feature Point Detection Based on Local Neighbor Pixels
QIN Hua-Biao, ZHANG Ya-Ning, CAI Jing-Jing
To solve the problem of low computational efficiency and many redundant feature points in feature point detection algorithm, a fast spatial-temporal feature point detection algorithm based on local neighbor pixels is proposed. The spatial-temporal feature points are located quickly by finding the points with great difference in pixel value in 3D spatial-temporal local neighborhood. Then the redundant feature points are removed with the 3D non-maxima suppression method, and the screened feature points are applied to human action recognition. In addition, the range of pixel segmentation threshold in local area and other detection problems of parameter optimization are analyzed according to binomial probability distribution principle. The experimental results show that the proposed algorithm not only improves the speed of feature point detection but also reliably detects enough amounts of feature points with the least redundancy, which leads to the high accuracy in human action recognition.
2015 Vol. 28 (1): 74-79 [Abstract] ( 469 ) [HTML 1KB] [ PDF 555KB] ( 826 )
80 Artificial Bee Colony Algorithm with Good Point Set and Turn Process of Monkey Algorithm
LIU Xiang-Pin, XUAN Shi-Bin, LIU Feng
An improved Artificial Bee Colony (ABC) algorithm with good point set and turn process of Monkey algorithm (MA) is proposed to overcome the defections of the basic ABC algorithm. Firstly, aiming at the defects of premature convergence, the good point set is used to initialize the population, which can generate a homogeneous population to keep the diversity of a swarm. Besides, the turn process of MA is introduced to help the swarm to jump out of the local optima and to get the global optimal solution.Simulation results on standard test functions and benchmark functions of CEC05 show that the proposed algorithm outperforms the basic algorithm and other improved ABC algorithms on both the precision and the convergence rate.
2015 Vol. 28 (1): 80-89 [Abstract] ( 462 ) [HTML 1KB] [ PDF 952KB] ( 839 )
90 Visual Tracking Algorithm Combining ORB Feature and Color Model
ZHONG Hua-Min, WANG Wei, ZHANG Hui-Hua

To solve the problem of invalid tracking by traditional CAMShift owing to the background with similar colors, a dynamic visual tracking algorithm is proposed combining ORB feature and color model of the object. The ORB feature is applied to extract the initial position of the object, and a adaptive color-threshold segmentation algorithm is proposed to improve the accuracy of color model for the object. Besides, the information of ORB feature points is used to revise the search window in the tracking procedure, which improves the tracking accuracy and robustness. Furthermore, a new method is proposed to estimate whether the moving object is missing, and an iteratively updated feature template is built to relocate the disappeared target. The experiments on video sequence images demonstrate that the proposed algorithm outperforms CAMShift and other improved algorithms based on feature extraction. When the target moves at high speed, the proposed algorithm has good robustness and can find out the wrong tracking result and correct it. Moreover, the computational efficiency rises greatly to ensure the real-time performance.
2015 Vol. 28 (1): 90-96 [Abstract] ( 534 ) [HTML 1KB] [ PDF 2175KB] ( 1454 )
模式识别与人工智能
 

Supervised by
China Association for Science and Technology
Sponsored by
Chinese Association of Automation
NationalResearchCenter for Intelligent Computing System
Institute of Intelligent Machines, Chinese Academy of Sciences
Published by
Science Press
 
Copyright © 2010 Editorial Office of Pattern Recognition and Artificial Intelligence
Address: No.350 Shushanhu Road, Hefei, Anhui Province, P.R. China Tel: 0551-65591176 Fax:0551-65591176 Email: bjb@iim.ac.cn
Supported by Beijing Magtech 　Email:support@magtech.com.cn