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.10, Published 2017-10-31 Papers and Reports    Researches and Applications        Papers and Reports 865 Image Aesthetic Quality Scoring Method Based on Feature Complementation XIE Yanjuan, CHEN Zhaojiong, YE Dongyi The deep convolutional neural network method can hardly analyze specific regions of an image and the relationship between those regions. A method for image aesthetic quality assessment is proposed by means of complementary combination of deep and handcrafted features. The specific regions dominating the aesthetic quality of the image are identified. Then, five groups of aesthetic relevant handcrafted features including line angles feature and clarity comparison feature are selected and designed. The deep features are acquired using Siamese network. Support vector regression algorithm is then applied to evaluate the score of the aesthetic quality of the image based on those handcrafted and deep features. The score is adjusted and finalized in light of the weight of Spearman rank-order correlation coefficient. Experimental results show that the proposed method outperforms the existing methods and the result is consistent with subjective assessment results. 2017 Vol. 30 (10): 865-874 [Abstract] ( 713 ) [HTML 1KB] [ PDF 766KB] ( 628 ) 875 Object Tracking Algorithm Based on Object Saliency and Adaptive Background Constraint WANG Jing, ZHU Hong To seek a solution of tracking drift and object loss resulted from environmental interference and appearance change of the object, a object tracking algorithm via object saliency and adaptive background constraint is proposed. In the tracking framework of particle filter, the pixel characteristics of the object and the extended object are firstly weighted to construct the explicit model of the object according to the principle of Bayesian saliency. Next, the background around the object is considered adaptively by exploiting the saliency of the background. Finally, by judging the current appearance state of the object, the tracking result is obtained by taking advantage of the correlation between the object and the background. Matching error is reduced by the object saliency model, while tracking accuracy is improved by the adaptive constraint of background with occluded object and changed pose. The experimental results demonstrate the proposed method with stronger robustness and higher precision for object tracking. 2017 Vol. 30 (10): 875-884 [Abstract] ( 501 ) [HTML 1KB] [ PDF 2802KB] ( 744 ) 885 Discriminatively Trained Action Recognition Model Based on Hierarchical Part Tree QIAN Yinzhong, SHEN Yifan Action recognition of body pose from static image is exploited in this paper. A hierarchical part tree structure is proposed. In the structure, each node is composed by a collection of poselets to represent its pose variation and pairs of linked nodes are constrained to form a pictorial structure. Grounded on the structure, a discriminatively trained action recognition model based on hierarchical part tree is presented. Except for deforming cost, the pairwise potential function in the model introduces co-occurrence cost. Parent part contains child part and the relative position of linked nodes is described by normal distribution, and thus the matching procedure is inferred efficiently in the framework of distance transform and message passing. Three models with different number of nodes by trimming the tree are comparatively evaluated on two datasets. Experimental results demonstrate that coarse parts in former two layers have strong saliency for action recognition, the recognition capability is further improved by body parts in the third layer, and the anatomical stick parts in the fourth layer are basically not useful for action recognition. 2017 Vol. 30 (10): 885-893 [Abstract] ( 488 ) [HTML 1KB] [ PDF 1136KB] ( 533 ) 894 Group Decision Making Model Based on Probabilistic Hesitant Fuzzy Information Aggregation Operations WU Wenying, LI Ying, JIN Feifei, NI Zhiwei, ZHU Xuhui Serious loss of the decision information is caused while the decision making information is described by the existing hesitant fuzzy sets. In this paper, a multi-attribute group decision making model is designed on the basis of the probabilistic hesitant fuzzy information aggregation operators. Firstly, the Archimedean norm is introduced under the probability hesitation fuzzy environment to define the probability hesitation fuzzy operation rule. Secondly, the generalized probabilistic hesitant fuzzy ordered weighted averaging(GPHFOWA) operator and the generalized probabilistic hesitant fuzzy ordered weighted geometric(GPHFOWG) operator are proposed based on the operation rule, and their basic properties are discussed. Subsequently, several common forms of GPHFOWA operator and GPHFOWG operator and their relations are analyzed. Finally, a probabilistic fuzzy multi-attribute group decision making model is constructed by using the two kinds of operators, and the feasibility and the effectiveness of the decision model are verified by an example of supplier selection. 2017 Vol. 30 (10): 894-906 [Abstract] ( 697 ) [HTML 1KB] [ PDF 962KB] ( 553 ) 907 Microblog User-Tag Recommendation Algorithm Based on Noise Reduction Relation Regularization LIU Huiting, GUO Xiaoxue, CHENG Lei, ZHAO Peng The existing microblog user-tag recommendation methods mostly rely on friends relationship or content to realize recommendation, and the bandwagon relationship existing in microblog (noise problem) can not be discovered and the user label sparse problem is not solved. Therefore, a microblog user-tag recommendation algorithm based on noise reduction relation regularization is presented. The similarity of the user′s and his friends′ interests is measured by the micro-blog theme of users extracted by LDA to reduce the influence of those friends without interests in common with the target user. The noise reduction relationship is taken as the regularization item and it is introduced into user-tag nonnegative matrix factorization model to solve the user-tag sparse problem. The model is optimized and constrained via the Lagrange multiplier method and the KKT conditions, and finally the approximate user-tag matrix for recommended users′ tag is obtained. The experimental results show the proposed method exposes the high quality in recommendation. 2017 Vol. 30 (10): 907-916 [Abstract] ( 468 ) [HTML 1KB] [ PDF 914KB] ( 424 ) Researches and Applications 917 Artists Recognition via Line Shape and Ink Color Distribution of the Principal Direction for Chinese Paintings LIU Shang, SHENG Jiachuan Most of the existing Chinese paintings research focuses on the content rather than the artistic style. Since the essence of Chinese paintings is represented by brushwork and ink, different artists can be normally identified by the style of their brushwork. In this paper, an algorithm is proposed to classify Chinese paintings based on brushwork and ink. The line shape feature and the ink color distribution feature are described. Combining these two features, a complex character is established. And the complex character is used as the input of the support vector machines classifier. Extensive experiments show that the average recall and precision of the proposed algorithm are higher than those of the representative existing benchmarks, including MHMM, C4.5 and Fusion. The proposed algorithm can be used for the digital analysis, management, understanding and identification of Chinese paintings. Moreover, it provides an effective digital tool for the inheritance and appreciation of Chinese painting. 2017 Vol. 30 (10): 917-927 [Abstract] ( 582 ) [HTML 1KB] [ PDF 23571KB] ( 492 ) 928 Unsupervised Feature Selection for Interval Ordered Information Systems YAN Yuejun, DAI Jianhua There are a number of unsupervised feature selection methods proposed for single-valued information systems, but little research focuses on unsupervised feature selection of interval-valued information systems.In this paper, a fuzzy dominance relation is proposed for interval ordered information systems. Then, fuzzy rank information entropy and fuzzy rank mutual information are extended to evaluate the importance of features. Consequently, an unsupervised feature selection method is designed based on an unsupervised maximum information and minimum redundancy(UmImR) criterion. In the UmImR criterion, the amount of information and redundancy are taken into account. Experimental results demonstrate the effectiveness of the proposed method. 2017 Vol. 30 (10): 928-936 [Abstract] ( 533 ) [HTML 1KB] [ PDF 707KB] ( 573 ) 937 Semi-supervised Neural Machine Translation Based on Sentence-Level BLEU Metric Data Selection YE Shaolin, GUO Wu The performance of statistical machine translation is improved by language model. However, the monolingual corpus is not equal to be effectively used by neural machine translation. To solve this problem, a semi-supervised neural machine translation model based on sentence-level bilingual evaluation understudy(BLEU) metric data selection is proposed. The candidate translations for non-labeled data are firstly generated by statistical machine translation and neural machine translation models, respectively. Then the candidate translations are selected through sentence-level BLEU, and the selected candidate translations are added to the labeled dataset to conduct semi-supervised joint training. The experimental results demonstrate the effectiveness of the proposed algorithm in the usage of non-labeled data. In the NIST Chinese-English translation tasks, the proposed method obtains an obvious improvement over the baseline system only with the fine labeled data. 2017 Vol. 30 (10): 937-942 [Abstract] ( 545 ) [HTML 1KB] [ PDF 557KB] ( 444 ) 943 Locality Constraint Enhanced Least Squares Regression Subspace Clustering ZHAO Jian, WU Xiaojun, DONG Wenhua Least square regression subspace clustering(LSR) is the lack of local correlation information of data, and thus dense representation is caused. Aiming at this problem, locality constraint enhanced least squares regression subspace clustering(LC_LSR) is proposed. The original algorithm of LSR is extended by adding the local correlation constraint to achieve an accurate coefficient matrix and then it is close to being block diagonal. Furthermore, a method to construct affinity matrix is proposed. The proposed algorithm can better strengthen the affinities within each cluster and weaken the ones across clusters. Experimental results show that the proposed algorithm effectively improves the accuracy of clustering and its effectiveness and feasibility are verified. 2017 Vol. 30 (10): 943-951 [Abstract] ( 608 ) [HTML 1KB] [ PDF 1122KB] ( 483 ) 952 Fuzzy Rough Set Model Based on Label Relations GUO Rongchao, LI Deyu, WANG Suge The data in multi-label classification tasks are usually high dimensional. Utilizing high-dimension data directly for modeling often results in a lower training efficiency or a complex model with the classifier performance reduced. For multi-label data, the concept of attribute-label matrix is proposed, a label relation based fuzzy rough set model is established, and a reduction algorithm of the model is then designed for feature selection of multi-label classification tasks. Finally, the effectiveness of the proposed method is verified on eight public datasets. 2017 Vol. 30 (10): 952-960 [Abstract] ( 527 ) [HTML 1KB] [ PDF 542KB] ( 911 )

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