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....

2015 Vol.28 Issue.9, Published 2015-09-30 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 769 Approximate Reasoning Model Based on Probability Valuation of Propositional Logic ZHANG Jia-Lu , CHEN Xue-Gang , WU Xia The value domain of proposition logic is extended from two values {0,1} to a probability space, and hence the concept of probability valuation of propositional formulas is introduced. Probability valuation is a generalization of classical propositional valuation and various truth degrees. Based on probability valuation, the concepts of probability truth degree, uncertainty degree, probability truth degree based on the set of all probability valuation of formulas on independent events are introduced. Grounded on the discussion of the properties of probability truth degree, probability truth degree satisfies Kolmogorov axioms on the entire set of propositional formulas. It is proved that the set of probability truth degrees of all formulas based on the set of all probability valuation on independent events has no isolated points in [0,1]. In the form of deduction in propositional logic, the uncertainty degree of conclusion is less than or equal to the sum of the product of uncertainty degree of each premise and its essentialness degree in a formal inference. Based on probability valuation, some concepts of a.e.conclusion, conclusion in probability and conclusion in probability truth of a formula set are introduced, and the relations between these concepts are discussed. Moreover, two different approximate reasoning models based on probability valuation are proposed. 2015 Vol. 28 (9): 769-780 [Abstract] ( 528 ) [HTML 1KB] [ PDF 424KB] ( 646 ) 781 Multi-class Classification Algorithm Based on Ensemble Learning and Hierarchical Structure ZOU Quan , SONG Li , CHEN Wen-Qiang , ZENG Jian-Cang , LIN Chen The classification algorithm is an important research field in data mining and pattern recognition. A multi-class classification algorithm based on ensemble learning and hierarchical structure is proposed. Firstly, the problems are decomposed according to their hierarchy categories. The hierarchical structure of the hierarchical classifier is defined. Then, multiple weak classifiers are integrated by ensemble learning methods based on the hierarchical structure. Thus, the classification process is completed. In the data mining competition of CCDM 2014, the proposed algorithm achieves the highest performance on several indexes, including average accuracy and F1-score. The results verify the feasibility on the classification problem. 2015 Vol. 28 (9): 781-787 [Abstract] ( 576 ) [HTML 1KB] [ PDF 484KB] ( 875 ) 788 Vector Quantization Landmark Points for Supervised Isometric Mapping with Explicit Mapping CHEN Shi-Wen, WANG Xian-Bao, LI Meng-Yuan, YAO Ming-Hai Since isometric mapping (ISOMAP) has no supervision and explicit mapping function and other limitations, an improved algorithm, selection of vector quantization landmark points for supervised isometric mapping with explicit mapping (SE-VQ-ISOMAP), is put forward. Firstly, the category information is introduced in the construction of neighborhood graph and geodesic distance matrix. Aiming at the problem that the landmark points are introduced into iterative optimization when distance matrix is processed, a method of vector quantization is employed instead of the traditional random selection. Thus, the whole manifold structure is indicated better by the selected samples. Finally, the radial function is regarded as basis, and consequently explicit mapping of dimensionality reduction method is obtained. On the handwritten digits sets and UCI datasets, the experimental results show that the proposed algorithm is fast and stable with a higher recognition rate. 2015 Vol. 28 (9): 788-794 [Abstract] ( 446 ) [HTML 1KB] [ PDF 581KB] ( 739 ) 795 Fast Attribute Reduction Algorithm Based on Row Storage LIANG Bao-Hua, WANG Shi-Yi The existing attribute reduction algorithms mainly focus on the area of resident data in the memory. To decrease the accessing disk I/O times, a row storage mode is proposed. In this mode, not all data are required storing in the main memory. During the reducing process, the sub divisions of same category are collected into one array to get the simplified decision table quickly. Meanwhile, the indiscernibility degree is introduced as the measurement of the attribute importance. Then, a fast attribute reduction algorithm is proposed. Its time complexity and space complexity are low. The examples and experimental results show the effectiveness and feasibility of the proposed algorithm. 2015 Vol. 28 (9): 795-801 [Abstract] ( 448 ) [HTML 1KB] [ PDF 383KB] ( 498 ) 802 Target User′s Neighbors Modification Based Collaborative Filtering ZHANG Jia, LIN Yao-Jin, LIN Meng-Lei, LIU Jing-Hua In user-based collaborative filtering algorithm, the nearest neighbors of the target user are not accurate and reliable due to the tendency of user′s rating and the sparsity of rating matrix. An effective algorithm is presented to obtain user′s nearest neighbors. Firstly, the definitions of positive and negative ratings for user group are given respectively, and the nearest neighbors of target user are selected from the group containing same rating tendency. Then, the nearest neighbors of target user with few common rating items and high similarity are corrected. Thus, the final nearest neighbor collection is obtained. Experimental results show that the modified algorithm of neighbor selection improves the recommended quality effectively to some extent. 2015 Vol. 28 (9): 802-810 [Abstract] ( 702 ) [HTML 1KB] [ PDF 522KB] ( 794 ) Surveys and Reviews 811 Facial Image Pattern Recognition Based on Triple Space Fusion GAO Xin-Bo , WANG Nan-Nan , PENG Chun-Lei , LI Cheng-Yuan With the help of human experience knowledge and cognition, the performance of pattern recognition can be improved for some complex applications, and therefore it is important to construct a pattern recognition system based on the fusion of the physical space, the cyberspace and the cognitive space. In this paper, facial image recognition, which is widely applied in forensic evidence, is taken as an example. Its recent advances in pattern recognition based on triple space fusion are summarized. Sketch based face recognition techniques are introduced from three aspects: synthesis based methods, common space projection based methods and feature descriptor based methods. Some discussions and further development directions are also given. These methods provide technical support for some applications in the field of public security. 2015 Vol. 28 (9): 811-821 [Abstract] ( 606 ) [HTML 1KB] [ PDF 561KB] ( 821 ) Researches and Applications 822 Drawing Style Recognition of Facial Sketch Based on Multiple Kernel Learning ZHANG Ming-Jin , LI Jie , WANG Nan-Nan The drawing style recognition of facial sketches is widely used for painting authentication and criminal investigation. A drawing style recognition algorithm of facial sketch based on multiple kernel learning is presented. Firstly, according to the way of art critics recognize the drawing style of facial sketch, five parts, the face part, left eye part, right eye part, nose part and mouth part, are extracted from the facial sketch. Then, gray histogram feature, gray moment feature, speeded-up robust feature and multiscale local binary pattern feature are extracted from each part on the basis of artists′ different understandings of lights and shadows on a face and various usages of the pencil . Finally, different parts and features are integrated and the drawing styles of facial sketches are classified by multiple kernel learning. Experimental results demonstrate that the proposed algorithm has better performance and obtains higher recognition rates. 2015 Vol. 28 (9): 822-827 [Abstract] ( 671 ) [HTML 1KB] [ PDF 631KB] ( 860 ) 828 Discovery of Overlapping and Hierarchical Communities Based on Extended Link Cluster Sequence GUO Hong, HUANG Jia-Xin, GUO Kun The mining and discovery of overlapping and hierarchical communities is a hot topic in the area of social network research. Firstly, an algorithm, discovery of link conmunities based on extended link cluster sequence (DLC_ECS), is proposed to detect overlapping and hierarchical communities in social networks efficiently. Based on the extended link cluster sequence corresponding to community structures with various densities, the optimal link community is detected after searching for the global optimal density. The link communities are transformed into the node communities, and thus the overlapping communities can be found out. Then, hierarchical link communities extraction based on extended link cluster sequence (HLCE_ECS) is designed. Hierarchical link communities from the extended link cluster sequence is found by the proposed algorithm. The link communities are transformed into the node communities to find out the overlapping and hierarchical communities. Experimental results on are artificial and real-world datasets demonstrate that DLC_ECS algorithm significantly improves the community quality and HLCE_ECS algorithm effectively discovers meaningful hierarchical communities. 2015 Vol. 28 (9): 828-838 [Abstract] ( 446 ) [HTML 1KB] [ PDF 635KB] ( 518 ) 839 Factors Evaluation of Fog-Haze Weather Based on Hesitant Fuzzy Preference Relations JIN Fei-Fei , NI Zhi-Wei Aiming at group decision making problems under hesitant fuzzy environment, the concept of hesitant fuzzy preference relation is introduced. The additive consistency, multiplicative consistency and order consistency of the hesitant fuzzy preference relations are defined. A decision-making method of ranking alternative method is established based on the additive consistency and multiplicative consistency to obtain the priority weights. The consistency of hesitant fuzzy preference relations is improved effectively by the proposed method, and thus the improved preference relations can satisfy the need of additive consistency and multiplicative consistency. The optimization models are further developed to get the priority weight vector of the alternatives. Meanwhile, the preference information of the decision maker is retained in the decision process as much as possible, and the model is simplified. Therefore, the application of the hesitant fuzzy theory is greatly extended. The practicability and the effectiveness of the proposed method are verified through the experiment of affecting factors of fog-haze weather. 2015 Vol. 28 (9): 839-847 [Abstract] ( 547 ) [HTML 1KB] [ PDF 389KB] ( 730 ) 848 Multipose Face Image Recognition Based on Image Synthesis WANG Ya-Nan, SU Jian-Bo Pose variations of face images make recognition rate decrease significantly. The fused face image gained by several multi-pose face images of the same person is used to identify the face information of this person. The proposed fusion method includes texture information and geometry information, and the face image sets for fusion are selected through geometry information of face images to guarantee the integrity of the face information. On the basis of existing face databases, the recognition rates of the original face images and the fused face images composed of multi-pose face images from internet are tested respectively. The recognition results show that the fusion method achieves higher recognition rate. 2015 Vol. 28 (9): 848-856 [Abstract] ( 574 ) [HTML 1KB] [ PDF 1275KB] ( 968 ) 857 Evidence Combination Method Based on Included Angle Cosine HU Jia-Ji, LI Xin-De, WANG Feng-Yu Dempster-Shafer theory of evidence combination is widely used in the field of information fusion. However, it has a counter intuitive problem in highly conflictive evidence combination. Aiming at this problem, the source of evidence is analyzed, and then an evidence combination method based on cosine is proposed. Firstly, Pignistic probability is utilized to analyze the evidence source. Next, conflictive evidence is identified by comparing the included angle cosine with the threshold. Conflictive evidence is revised according to the variable correction factor generated from the relationship among evidences and the probability assignment of non-conflictive evidences. Finally, Dempster-Shafer rule is utilized to combine evidences. Experimental results show that the proposed method is effective in handing highly conflictive evidences and the combination result obtained in the case of multi-conflictive evidences is good. The proposed method can be widely used in the field of multi-source information fusion, pattern recognition and uncertain information decision-making. 2015 Vol. 28 (9): 857-864 [Abstract] ( 424 ) [HTML 1KB] [ PDF 466KB] ( 566 )

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