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.5, Published 2017-05-31 Papers and Reports    Researches and Applications        Papers and Reports 385 Optimized Index Assignment with Bi-directional Prediction in Palette Coding SONG Chuanming, CHEN Guisheng, HE Xing, FU Bo, WANG Xianghai In palette coding, the research on influence of index assignment on the coding efficiency of index map is insufficient. An optimized index assignment algorithm is proposed in this paper. With the bidirectional feedback between the index map generation and prediction, the index assignment is optimized by forward prediction, backward prediction and joint probability maximization. Thus, it is easier to predict index map accurately, and the efficiency of palette coding is improved. Experimental results show that the proposed algorithm increases the prediction accuracy of the local directional prediction algorithm compared with the multi-stage prediction algorithm. 2017 Vol. 30 (5): 385-393 [Abstract] ( 479 ) [HTML 1KB] [ PDF 801KB] ( 456 ) 394 Double-Quantitative Integration of Three-Way Decisions and Three-Way Attentions ZHANG Xianyong, YANG Jilin, TANG Xiao The conditional probability used in three-way decisions only exhibits relativity, and thus the introduction and integration of absoluteness measures are useful for rule extraction. The absolute conditional probability is mined to establish three-way attentions, and the double-quantitative integration of three-way decisions and attentions is investigated. Firstly, the relative and absolute conditional probabilities are extracted, and their systematic relationships are analyzed to reveal their heterogeneity and complementarity. Then, three-way attentions are produced by the absolute conditional probability, and their double-quantitative integration with three-way decisions is made. Thus, the integrated region type and basic semantics (granule) system are achieved. Finally, a statistical decision table is provided for illustration. Based on the absolute conditional probability, three-way attentions become a new type of three-way patterns, and their double-quantitative integration with three-way decisions shows systematicness and applicability. 2017 Vol. 30 (5): 394-402 [Abstract] ( 440 ) [HTML 1KB] [ PDF 698KB] ( 484 ) 403 Tag Clustering Method of Joint Topic Model HU Xuegang, LI Huizong, PAN Jianhan, HE Wei, YANG Hengyu Improving the clustering quality of social tags is a key problem in the semantics recognition of tags. A joint topic model based on resource is proposed to cluster tags. Firstly, reference relations of the resource are utilized to acquire the authority scores of resource by using random walk method. Secondly, the resource authority is applied to set the weights of two binary relations of resource-tag and resource word. Grounded on that, the joint latent Dirichlet allocation(LDA) model of the word and the tag based on resource weighted is constructed. By iterative learning, the latent topics of the tag are acquired, and the clusters are decided according to the maximum membership degree of the tag. The results show that the proposed method has a better clustering performance than other tag clustering methods based on resource. 2017 Vol. 30 (5): 403-415 [Abstract] ( 521 ) [HTML 1KB] [ PDF 886KB] ( 420 ) 416 Feature Selection Algorithm Based on Neighborhood Valued Tolerance Relation Rough Set Model YAO Sheng, XU Feng, ZHAO Peng, WANG Jie, CHEN Ju The existing methods of feature selection are mostly based on tolerance relation in the numerical incomplete information system.However, the data similarity characterization is too loose in these approaches. Therefore, the rough set model of neighborhood valued tolerance relation is proposed in this paper. The neighborhood valued tolerance condition entropy is defined on the basis of the model. And the related properties are analyzed.Finally, the corresponding algorithm is constructed according to the monotonicity of neighborhood valued tolerance condition entropy. Experimental results show that the proposed algorithm is superior to the existing algorithms in terms of the feature selection results, arithmetic operation time and classification accuracy. 2017 Vol. 30 (5): 416-428 [Abstract] ( 528 ) [HTML 1KB] [ PDF 997KB] ( 554 ) 429 Informative Gene Selection Method Based on Symmetric Uncertainty and SVM Recursive Feature Elimination YE Mingquan, GAO Lingyun, WU Changrong, WAN Chunyuan A large number of genes unrelated to tumor classification exist in gene expression profiles, and thus the prediction accuracy of tumor is reduced substantially. Due to the small sample size with high dimension and noise, the tumor diagnosis is harder. To get an informative gene subset with fewer genes and a better classification accuracy, an informative gene selection method based on symmetric uncertainty(SU) and support vector machine-recursive feature elimination(SVM-RFE) is proposed. Firstly, SU is used to evaluate the correlation between genes and class labels, and approximate Markov blanket is defined grounded on SU. The elimination of irrelevant and redundant genes is achieved. Secondly, SVM-RFE algorithm is applied to obtain the effective informative gene subset by further removing redundant genes. Experimental results show that the proposed algorithm produces higher classification performance with equal or less informative gene subset. 2017 Vol. 30 (5): 429-438 [Abstract] ( 661 ) [HTML 1KB] [ PDF 803KB] ( 928 ) Researches and Applications 439 Clustering Algorithm with Kernel Density Estimation ZHU Jie, CHEN Lifei Similarity measure is an important basis for clustering analysis. However, defining an efficient similarity measure for discrete symbols (categories) is difficult. In this paper, a method is proposed to measure the similarity between categories in terms of their kernel probability density. Different from the traditional simple-matching method or frequency-estimation method, under the action of the bandwidth for kernel functions, the proposed measure no longer depends on the assumption that categories on the same attribute are statistically independent. Then, a Bayesian clustering model is established based on kernel density estimation of categories, and a clustering algorithm is derived to optimize the clustering model using a likelihood-based object-to-cluster similarity measure. Finally, three data-driven approaches are proposed by leave-one-out estimation and maximum likelihood estimation to dynamically determine the optimal bandwidths in the kernel function for clustering. Experiments are conducted on real-world datasets and the results demonstrate that the proposed algorithm achieves higher clustering accuracy compared with the existing algorithms using a simple-matching distance measure or the attribute-weighting variants. The results also show that the bandwidth estimated by the proposed algorithm has practical significance in the applications, such as important feature identification. 2017 Vol. 30 (5): 439-447 [Abstract] ( 659 ) [HTML 1KB] [ PDF 589KB] ( 576 ) 448 Multi-level Deep Network Fused for Face Recognition HU Zhengping, HE Wei, WANG Meng, SUN Zhe Discriminative facial features can be obtained by deep learning model. Therefore, combining the deep learning, a multi-level deep network extraction model for fusion feature is proposed. In the proposed model, the pooling layer is added after subspace mapping based on deep subspace model, so that the feature dimension is reduced with texture details preserving and local transformation robustness.Meanwhile, face region is divided into 5 parts according to facial feature point achieved by face alignment algorithm. Based on multi-level classification strategy, the global network is firstly trained using the whole face image to obtain five candidate labels for test sample. Then, the local face block is put into sub-network to obtain local representation and test samples are classified in the candidate labels. Experimental results show that the model combined with the local features and global features achieves better accuracy and robustness in the aspect of the illumination, expression, occlusion, etc. Moreover, adding pooling structure and the two-step discrimination algorithm effectively improve the recognition efficiency. 2017 Vol. 30 (5): 448-455 [Abstract] ( 737 ) [HTML 1KB] [ PDF 909KB] ( 767 ) 456 Fabric Defect Detection Based on Similarity Relation LIANG Jiuzhen, GU Chengxi, CHANG Xingzhi Focusing on the fabric defect detection with periodic variation pattern, a fabric defect detection method based on similarity relation is proposed. Firstly, the size of the periodic model is conformed. Secondly, grounded on the equivalence class partition method, block clustering is performed according to the cycle size (template). Then, the defect blocks are located. The similarity relation between blocks is transformed into equivalence relation and a threshold segmentation strategy is put forward. Finally, the defect detection method based on neighborhood information is added to complete the detection process. Experiments show that by the proposed method the detection accuracy is improved substantially, and the detection process is simpler and more practical. 2017 Vol. 30 (5): 456-464 [Abstract] ( 532 ) [HTML 1KB] [ PDF 948KB] ( 573 ) 465 Entity Relation Extraction Based on Convolutional Neural Network and Keywords Strategy WANG Linyu, WANG Li, ZHENG Tingyi The conventional relation extraction methods are time consuming, the error propagation in feature selection is likely to emerge, and the deep learning methods only depend on word embeddings to learn features. Aiming at these problems, a relation extraction method based on convolutional neural network and keywords strategy is proposed. Based on feature of the word embeddings, the keywords feature is acquired by the term proportion-inverse sentence proportion (TP-ISP) keywords extraction algorithm based on sentence. Thus, the category division is increased and the deficiency of the network to automatically learn features from sentence is remedied. In the network training process, the chunk-based max pooling strategy is adopted to reduce the information loss by the traditional max-over-time pooling strategy. The experiment demonstrates that the proposed method improves the results of entity relation extraction. 2017 Vol. 30 (5): 465-472 [Abstract] ( 774 ) [HTML 1KB] [ PDF 652KB] ( 1265 ) 473 Knowledge Graph Reasoning Based on Paths of Tensor Factorization WU Yunbing, ZHU Danhong, LIAO Xiangwen, ZHANG Dong, LIN Kaibiao In the existing tensor factorization techniques used in knowledge graph learning and reasoning, only direct links between entities are taken into account. However, the graph structure of knowledge graph is ignored. In this paper, knowledge graph reasoning based on paths of tensor factorization is proposed. The path ranking algorithm(PRA) is employed to find all paths connecting the source and target nodes in a relation instances. Then, those paths are decomposed by tensor factorization. And the entities and relations are optimized by the alternating least squares method. Experimental results on two large-scale knowledge graphs show the algorithm achieves significant and consistent improvement on tasks of entities linking prediction and paths question answering and its prediction accuracy outperforms that of other related models. 2017 Vol. 30 (5): 473-480 [Abstract] ( 850 ) [HTML 1KB] [ PDF 714KB] ( 1980 )

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