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....

2020 Vol.33 Issue.8, Published 2020-08-25 Papers and Reports    Researches and Applications        Papers and Reports 671 Automatic Colorization Algorithm with Anime Effect for Scene Sketches ZHU Song, CHEN Zhaojiong, YE Dongyi When the existing automatic portrait coloring algorithms are directly applied to scene sketches, distortion phenomena are caused, such as wrong colorization and checkerboard artifacts,due to the diversified line semantics of scene sketches. To address this issue, an automatic colorization algorithm with anime effect for scene sketches is put forward. The structure of U-Net generator in the existing automatic portrait coloring algorithms is improved and enhanced based on the conditional generative adversarial network. A double-layer information extraction U-Net(DIEU-Net) is designed for automatic anime effect colorization of scene sketches. Firstly, the double-convolution sub-module prominence-information extraction of a scene sketch(IESS) is designed. Then, a module integrating double-layer IESS and residual structure is inserted into different stages of the proposed generator. Thus, the global learning ability of the generator on important features, like colors and positions related to the sketch, are enhanced, and the network degradation problems caused by vanishing gradients as the network deepens, are alleviated. Moreover, the deconvolution in U-Net is replaced by the operations of convolution and upsample to suppress the occurrence of the checkerboard artifacts. Experimental results show that the proposed algorithm performs well in avoiding the distortion phenomenon and achieves more reasonable and natural coloring effect than other algorithms. Furthermore, the proposed algorithm can be applied to automatic anime coloring of various types of scene sketches. 2020 Vol. 33 (8): 671-680 [Abstract] ( 727 ) [HTML 1KB] [ PDF 4179KB] ( 521 ) 681 Binocular Stereo Matching Algorithm Based on Labeled Matching Region Correction ZHOU Jiali, CHEN Yu, WU Chao, WU Min Binocular stereo vision is an accurate and effective measurement method. A binocular stereo matching method for label matching region correction is proposed. Based on the conventional graph cuts algorithm, the matching region is corrected by the spatial geometrical information of the block and a higher subpixel accuracy matching disparity map is obtained. Firstly, the correction transformation is determined by label and spatial geometrical information, and consequently the pixel information of the matching area of the left and right images are fully exploited. Then, the candidate 3D label is updated repeatedly to find the label minimizing the global energy. Finally, the left-right checking and the mean filtering are utilized to refine the disparity map. Experiments show that the proposed method is effective in finding a good, smooth piecewise linear disparity map with higher accuracy for edge region and occlusion region. 2020 Vol. 33 (8): 681-691 [Abstract] ( 512 ) [HTML 1KB] [ PDF 3492KB] ( 622 ) 692 Overlapped Features Strategy and Parameters Optimization Patterns Recognition for Motor Imagery EEG LUO Tianjian, ZHOU Changle Aiming at nonlinear and non-stationary characteristics of motor imagery(MI) based electroencephalogram, a features extraction and patterns recognition algorithm is proposed. Firstly, overlapped filter bank(OFB) pre-processing is conducted. Then, the common spatial patterns(CSP) algorithm is applied to the filtered electroencephalogram(EEG) signals. Afterwards, the OFB-CSP features are incorporated into robust support matrix machine(RSMM) for MI patterns recognition, and the corrected particle swarm optimization(CPSO) algorithm is utilized to dynamically adjust the optimal parameters for RSMM classification. Experiments on two public datasets show that OFB pre-processing improves the discrimination of CSP features. Besides, the optimal parameters for EEG signals of individuals are searched by CSPO to the RSMM classifier. Compared with the state-of-the-arts algorithms, the proposed algorithm significantly improves MI classification accuracy. With less requirements of samples and computational resources, the proposed overlapped features strategy and parameters optimization algorithm is suitable for real-world brain computer interface application. 2020 Vol. 33 (8): 692-704 [Abstract] ( 372 ) [HTML 1KB] [ PDF 810KB] ( 284 ) 705 Multi-label Label-Specific Features Learning Combined with Multi-category Correlation Information WU Anqi, GAO Qingwei, SUN Dong, LU Yixiang When the specific features of labels are extracted by most of the existing label-specific features learning methods, only the correlations among labels are taken into account and the correlations among instances and among features are neglected. Therefore, the classification accuracy are reduced. To solve this problem, an algorithm for multi-label label-specific features learning combined with multi-category correlation information is designed in this paper. Label correlation, feature correlation and instance correlation are considered. The label correlation between labels are calculated by cosine similarity. The similarity graph matrix is constructed to learn feature correlation and instance correlation. The specific features of labels are selected by the proposed algorithm compactly, the classification accuracy is improved and the problem of excessive dimensionality in multi-label classification is effectively solved. 2020 Vol. 33 (8): 705-715 [Abstract] ( 502 ) [HTML 1KB] [ PDF 790KB] ( 386 ) Researches and Applications 716 Feature Selection Algorithm Based on Label Correlation LÜ Yuejiao, LI Deyu In multi-label classification, each sample can be associated with multiple label classes at one time and some of them are related to each other. The classification performance is optimized by taking full advantage of these label correlations. Therefore, frequent itemsets are employed to mine the correlation between labels, and an improved multi-label feature selection algorithm is proposed for the multi-label attribute reduction algorithm based on neighborhood rough set. Then, the samples are further clustered and grouped according to the similarity of the features, and attribute reduction and classification are performed based on the label correlations in local samples. Finally, the effectiveness of the proposed algorithm is verified by experiments on 5 multi-label datasets. 2020 Vol. 33 (8): 716-723 [Abstract] ( 539 ) [HTML 1KB] [ PDF 517KB] ( 394 ) 724 Robust Uncertainty Measurement for Interval-Valued Decision Information System via Information Structure WU Yiyang, DAI Jianhua, CHEN Jiaolong Uncertainty measurement for single valued information system is widely studied. There are few researches on uncertainty measurement for interval-valued decision information system and the influence of the noise label on uncertainty measurement. Therefore, a robust uncertainty measurement for interval-valued decision information system via information structure is proposed. Firstly, the similarity degree between interval values is defined by KL divergence, and the fuzzy similarity relation of the interval values is constructed. Then, a information structure for interval-valued decision information system is proposed. In addition, K nearest neighbor points algorithm is introduced to calculate the membership degree of the samples about the decision, and two information structure based robust uncertainty measurement approaches are proposed to reduce the impact of noise labels on uncertainty measurement of systems. Finally, the validity and rationality of the proposed uncertainty measurement are verified through the experiments. 2020 Vol. 33 (8): 724-731 [Abstract] ( 379 ) [HTML 1KB] [ PDF 878KB] ( 329 ) 732 Cost-Sensitive Text Sentiment Analysis Based on Sequential Three-Way Decision FAN Qin, LIU Dun, YE Xiaoqing To solve the problems of cost imbalance in text sentiment analysis and high classification cost in static decision-making, a cost-sensitive text sentiment analysis method is constructed based on sequential three-way decision, and the misclassification cost and learning cost in dynamic decision-making process are taken into account. Firstly, a granulation model for text data is proposed to construct a multi-level granular structure. Next, sequential three-way decision is introduced to set a dynamic text analysis framework. Finally, real text review datasets are utilized to validate the effectiveness of the proposed method. Experimental results show that the proposed method significantly reduces the overall decision-making cost with the improved classification quality. 2020 Vol. 33 (8): 732-742 [Abstract] ( 379 ) [HTML 1KB] [ PDF 682KB] ( 426 ) 743 Sequential Three-Way Sentiment Analysis Based on Temporal-Spatial Multi-granularity YANG Xin, LIU Dun, LI Qiuke, YANG Xibei The traditional static methods of sentiment analysis cannot meet the quantity and complexity requirements of dynamic data in the big data era. Therefore, grounded on the concept of sequential three-way decisions, a sequential three-way sentiment analysis framework based on temporal-spatial multi-granularity is proposed to overcome the shortcomings of the traditional two-way decisions. Firstly, a multi-layer granular structure with temporal-spatial features is constructed using increasing data and better-fitting feature space over time to balance the misclassification cost and training cost. Then, three typical sentiment classification methods are applied as benchmarks to test the efficiency of the proposed method. Finally, compared with the static methods, experimental results on two datasets show that the proposed method greatly reduces the classification costs with the classification quality maintained. 2020 Vol. 33 (8): 743-752 [Abstract] ( 464 ) [HTML 1KB] [ PDF 703KB] ( 309 ) 753 Context-Oriented Attention Joint Learning Network for Aspect-Level Sentiment Classification YANG Yuting, FENG Lin, DAI Leichao, SU Han To solve the problems of weak perception for aspect words and generalization ability in the existing models for sentiment classification, a context-oriented attention joint learning network for aspect-level sentiment classification(CAJLN) is proposed. Firstly, the bidirectional encoder representation from transformers(BERT) model is employed as the encoder to preprocess short texts into sentences, sentence pairs and aspect words as input, and their hidden features are extracted through the single sentence and sentence pair classification models, respectively. Then, based on the hidden features of sentences and aspect words, attention mechanisms for sentences and aspect words are established to obtain aspect-specific context-aware representation. Then, the hidden features of sentence pairs and aspect-specific context-aware representations are learned jointly. Xavier normal distribution is utilized to initialize the weights. Thus, the continuous updating of the parameters during the back propagation is ensured, and useful information is learned by CAJLN in the training process. Experiments show that CAJLN effectively improves the performance of sentiment classification for short text on multiple datasets. 2020 Vol. 33 (8): 753-765 [Abstract] ( 600 ) [HTML 1KB] [ PDF 933KB] ( 543 )

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