Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文

Judgement and Disposal of Academic Misconduct Article Copyright Transfer Agreement Proof of Confidentiality Requirements for Electronic Version More.... Chinese Association of Automation National ResearchCenter for Intelligent Computing System Institute of Intelligent Machines,Chinese Academy of Sciences More....

2017 Vol.30 Issue.7, Published 2017-07-31 Orignal Article        Orignal Article 577 Parallel Imaging: A New Theoretical Framework for Image Generation WANG Kunfeng, LU Yue, WANG Yutong, XIONG Ziwei, WANG Fei-Yue To build computer vision systems with good generalization ability, large-scale, diversified, and annotated image data are required for learning and evaluating the in-hand computer vision models. Since it is difficult to obtain satisfying image data from real scenes, a new theoretical framework for image generation is proposed, which is called parallel imaging. The core component of parallel imaging is various software-defined artificial imaging systems. Artificial imaging systems receive small-scale image data collected from real scenes, and then generate large amounts of artificial image data. In this paper, the realization methods of parallel imaging are summarized, including graphics rendering, image style transfer, generative models, etc. Furthermore, the characteristics of artificial images and actual images are analyzed and the domain adaptation strategies are discussed. 2017 Vol. 30 (7): 577-587 [Abstract] ( 926 ) [HTML 1KB] [ PDF 4273KB] ( 1003 ) 588 Pedestrian Detection on Videos Based on Graph Cuts and Density Clustering ZENG Chengbin, LIU Jiqian In the existing pedestrian detection algorithms, the pedestrian detection is considered as a supervised learning problem of two classes, pedestrian and background. Thus, the pedestrian and the background in the video are distinguished. However, the problem of variable poses and heavy occlusion can not be solved by these algorithms effectively. In this paper, a pedestrian detection algorithm based on graph cuts and density clustering is proposed. The pedestrian detection is regarded as an unsupervised learning problem of multiple classes. At the training stage, the multilevel histogram of oriented gradient-local binary pattern(HOG-LBP) features are firstly calculated for each of training samples. Then, different weights are assigned to each image block of the multilevel HOG-LBP features. To distinguish the different parts of pedestrian and assign weight, the image sample is segmented by the block-based graph cuts algorithm. Finally, the density clustering approach is used to classify the positive and negative samples into multiple cluster center respectively. At the testing stage, the distance between the multilevel HOG-LBP of test sample and every cluster center is calculated, and the five shortest distances are voted to classify the test sample. Experiments show that the proposed algorithm can handle the pose variations and partial occlusions effectively. Moreover, with the increase of training samples, the results of the proposed algorithm can be comparable to that of the state-of-the-art pedestrian detection algorithms. 2017 Vol. 30 (7): 588-587 [Abstract] ( 701 ) [HTML 1KB] [ PDF 1400KB] ( 799 ) 598 Cross-Entropy Semi-supervised Clustering Based on Pairwise Constraints LI Chaoming, XU Shengbing, HAO Zhifeng The objective function used in the classical maximum entropy clustering(MEC) lacks the information expression on pairwise constraints. Therefore, the effective supervision information is wasted when a small amount of pairwise constraints are known. In this paper, an algorithm of cross-entropy semi-supervised clustering(CE-sSC) based on pairwise constrains on the basis of MEC algorithm is proposed. The sample cross-entropy is utilized to describe the pairwise constraints information and introduced to the objective function of MEC as a penalty term. With Lagrange optimization procedure, the objective function can be resolved into the cluster center and the membership update equations. Experimental results indicate the proposed method effectively improves the clustering performance by using a small amount of pairwise constraints and works well on actual datasets. 2017 Vol. 30 (7): 598-608 [Abstract] ( 899 ) [HTML 1KB] [ PDF 976KB] ( 1072 ) 609 Overlapping Subspace Clustering Based on Probabilistic Model QIU Yunfei, FEI Bowen, LIU Daqian Due to the low clustering accuracy of the existing subspace clustering methods in dealing with the problem of overlapping clusters, an overlapping subspace clustering algorithm based on probability model(OSCPM) is proposed. Firstly, the high-dimensional data is divided into several subspaces by using the subspace representation of mixed-norm. Then, a probability model of the exponential family distribution is used to determine the overlapping part of the clusters in the subspace, and the data is assigned to the correct class clusters to get the clustering results. An alternating maximization method is used to determine the optimal solution of the objective function in the process of parameter estimation. Experimental results on artificial datasets and UCI datasets show that OSCPM produces better clustering performance compared with other algorithms and it is suitable for large scale datasets. 2017 Vol. 30 (7): 609-621 [Abstract] ( 621 ) [HTML 1KB] [ PDF 891KB] ( 476 ) 622 Finger-Knuckle-Print Recognition: A Preliminary Review LU Jingting, JIA Wei, YE Hui, ZHAO Yang, MIN Hai, YU Ye, HU Rongxiang Compared with face, fingerprint, and iris based biometrics systems, finger-knuckle-print recognition based biometrics system has stable features, and it can be collected by low cost device and be easily combined with palmprint, finger vein, and hand shape recognition to form a robust system. In this paper, the definition, the data acquisition and the preprocessing of finger-knuckle-print recognition are firstly introduced. Then, the feature extraction and matching algorithms as well as multi-modal methods are reviewed. The effective finger-knuckle-print recognition algorithms are roughly divided into six categories: texture-based algorithm, structure-based algorithm, subspace learning-based algorithm, correlation filter-based algorithm, local descriptor-based algorithm and orientation coding-based algorithm. Finally, the development tendency of finger-knuckle-print recognition is forecasted. 2017 Vol. 30 (7): 622-636 [Abstract] ( 850 ) [HTML 1KB] [ PDF 797KB] ( 773 ) 637 Interactive Dimension Reordering in RadViz with Correlation Matrix ZHANG Zhihao, ZHANG Junping, CHAN Takming, LU Ying, YUAN Xiaoru, GU Tianlong In the existing dimension reordering algorithms, the interactive functions allowing users to perform order navigation or reorder operations are rarely taken into account. Aiming at this problem, a navigation-guided correlation matrix is proposed for users to interactively reorder dimensions in radial visualization(RadViz). A hierarchical clustering algorithm with configurable parameters is specially designed for RadViz to recommend the initial dimension order. The dendrogram of results is employed to help users interactively reorder,select and delete dimensions for feature subset selection. Experiments show that the proposed method is interactive, user-friendly and helpful for alleviating the overlapping problem of data projections in RadViz. 2017 Vol. 30 (7): 637-645 [Abstract] ( 622 ) [HTML 1KB] [ PDF 2341KB] ( 546 ) 646 Semi-supervised Ensemble Learning Based Software Defect Prediction WANG Tiejian, WU Fei, JING Xiaoyuan The software defect prediction is usually adversely affected by the limitation of the labeled modules and the class-imbalance of software defect data. Aiming at this problem, a semi-supervised ensemble learning software defect prediction approach is proposed. High-performance classifiers can be built through semi-supervised ensemble learning by using a large amount of unlabeled modules and a better prediction capability is achieved for class-imbalanced data by using a series of weak classifiers to reduce the bias generated by the majority class. With the consideration of the cost of risk in software defect prediction, a sample weight vector updating strategy is employed to reduce the cost of risk caused by misclassifying defective modules as non-defective ones. Experimental results on NASA MDP datasets show better software defect prediction capability of the proposed approach. 2017 Vol. 30 (7): 646-652 [Abstract] ( 727 ) [HTML 1KB] [ PDF 638KB] ( 637 ) 653 Sparse Feature Extraction Model Based on Deep and Symmetric Subspace Learning HU Zhengping, CHEN Junling, WANG Meng, SUN Zhe An algorithm of sparse feature extraction model is proposed based on deep and symmetric subspace learning. According to the theory of deep subspace learning, the constraints of symmetry and sparsity are introduced and the deep map network is built to extract features. Firstly, the basic subspace mapping matrix is constructed by minimizing the reconstruction error and the constraints of symmetry and sparsity are introduced for training. Next, the basic subspace model based on deep learning is reformed, and the deep symmetric sparse feature extraction model is built. These feature extraction results from different layers are merged to obtain the multi-layered deep symmetric subspace sparse feature. The experimental results on face databases show that the proposed algorithm achieves high recognition rates and strong robustness in illumination, expression and pose. Furthermore, compared with the convolutional neural networks, the proposed algorithm has the advantages of simple structure and high convergent rate. 2017 Vol. 30 (7): 653-662 [Abstract] ( 482 ) [HTML 1KB] [ PDF 862KB] ( 497 ) 663 Residual Value Iteration Algorithm Based on Function Approximation CHEN Jianping, HU Wen, FU Qiming Aiming at the problem of unstable and slow convergence of traditional value iteration algorithm, an improved residual value iteration algorithm based on function approximation is proposed. The traditional value iteration algorithm and the value iteration algorithm with Bellman residual are combined. Weight factors are introduced and new rules are constructed to update value function parameter vector. Theoretically, the new parameter vector can guarantee the convergence of the algorithm and solve the unstable convergence problem in the traditional value iteration algorithm. Moreover, the forgotten factor is introduced to speed up the convergence of the algorithm. The experimental results of Grid World problem show that the proposed algorithm has good performance and robustness. 2017 Vol. 30 (7): 663-672 [Abstract] ( 564 ) [HTML 1KB] [ PDF 623KB] ( 514 )

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