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.3, Published 2015-03-30 Papers and Reports    Researches and Applications        Papers and Reports 193 A Hybrid Discriminative Approach with Bayesian Prior Constraint YAO Ting-Ting, XIE Zhao, ZHANG Jun, GAO Jun The discriminative models are sensitive to limited training samples, which usually has poor generalization performances and is easily over-fitting. A hybrid discriminative approach with Bayesian prior constraints is proposed to solve this issue.By introducing the generative prior analysis into the discriminative approach, a complementary learning structure is built to fuse different classification results.The different types of classifiers are trained separately, and an effective fusion decision is defined to obtain the most confident testing samples along with the estimated labels. By enlarging the training set automatically, the model is updated to make up for the incomplete distribution information of training samples.The experimental results show that compared with the classical methods, the proposed approach can effectively update the model by figuring out the discriminating samples and correct the misclassifications caused by the uneven distribution of limited samples. It can improve the performances of scene categorization. 2015 Vol. 28 (3): 193-201 [Abstract] ( 665 ) [HTML 1KB] [ PDF 1238KB] ( 723 ) 202 λ-Resolution of Fuzzy Propositional Logic System with Three Kinds of Negation FLCOM ZHAO Jie-Xin, PAN Zheng-Hua Since the importance of automated reasoning and the resolution principle of the fuzzy logic with one negation is mainly studied now, the resolution principle of the fuzzy proposition logic(FLCOM) with three kinds of negation, contradictory negation, opposite negation and medium negation, is discussed. Based on an infinite-valued semantic interpretation of FLCOM, λ-satisfiable and λ-unsatisfiable concepts are proposed, and λ-resolution method is introduced into FLCOM. Besides, λ-resolution deduction of FLCOM is defined and λ-resolution principle of FLCOM is discussed. Moreover, the completeness of λ-resolution method is proved. Based on λ-resolution method and the proved conclusions, some examples providing evidences for the λ-resolution method and the conclusions are listed below the corresponding definitions and theorems. Therefore, whether a fuzzy propositional formula is λ-satisfiable or λ-unsatisfiable can be determined in the range of FLCOM. 2015 Vol. 28 (3): 202-208 [Abstract] ( 493 ) [HTML 1KB] [ PDF 365KB] ( 503 ) 209 Speech Recognition Based on Deep Neural Networks on Tibetan Corpus YUAN Sheng-Long, GUO Wu, DAI Li-Rong Large vocabulary continuous speech recognition on telephonic conversational Tibetan is firstly addressed in this paper. As a minority language, the major difficulty in Tibetan speech recognition is data deficiency. In this paper, the acoustic model of Tibetan is trained based on deep neural networks (DNN).To address the issue of data deficiencies, the DNN models of other majority languages are used as the initial networks of the objective Tibetan DNN model. In addition, phonetic questions of Tibetan generated by phonetic expert are unavailable due to the lacking knowledge of phonetics. To reduce the number of tri-phone hidden Markov models(HMM) in Tibetan speech recognition, phonetic questions automatically generated in the data driven manner are used for tying the tri-phone HMM. In this paper, different clustering of tri-phone states is tested and the words accuracy is about 30.86% on the test corpus by Gaussian mixture model(GMM). When the acoustic model is trained based on DNN, 3 kinds of DNN model trained by different large corpus are adopted. The experimental results show that the proposed methods can improve the recognition performance, and the words accuracy is about 43.26% on the test corpus. 2015 Vol. 28 (3): 209-213 [Abstract] ( 675 ) [HTML 1KB] [ PDF 347KB] ( 1210 ) 214 An Adaptive C-V Image Segmentation Model Guided by Gray Difference Energy Function WANG Xiang-Hai, WANG Jin-Ling , FANG Ling-Ling As the sign of geometric active contour model (GACM), the C-V model has robustness to obscured targets and edge noise in image segmentation. However, this model usually cannot deal with complex heterogeneous images, and it is also sensitive to the initial position of evolution curve and has a high computational complexity. The more the average gray difference between inner region and outer region is, the closer the evolutionary curve to accurate target edge is. On this basis, an adaptive C-V image segmentation model guided by gray difference energy function is proposed. The model can adjust the movement trend of the evolutionary curve by the guidance function constructed based on average gray difference between inner region and outer region adaptively. This makes the evolution of the curve within a valid narrow band scope. The proposed model ensures the local homogeneity of gray calculation of contour curve between inner region and outer region and enhances the ability to capture the detailed target. At the same time, it improves the calculation speed of the model and the adaptability to the initial position of evolution curve to a certain extent. A large number of simulation experiments verify the validity of the proposed model. 2015 Vol. 28 (3): 214-222 [Abstract] ( 517 ) [HTML 1KB] [ PDF 1632KB] ( 724 ) 223 Object Tracking via Multi-task Least-Soft-Threshold Squares Regression YASIN Musa, ZHAO Chun-Xia In visual object tracking, model representation is one of the core issues directly affecting tracking efficiency. The object model representation needs an efficient template learning from the online complicated data to adapt to the variations caused by intrinsic or extrinsic factors during the tracking process. In this paper, the detailed descriptions of robust object model representation algorithm is provided, and a novel multi-task least-soft-threshold squares tracking framework (MLST) is proposed. In the proposed scheme, the observation model is treated as a multi-task linear regression problem,and the appearance models of the candidate target under different states are represented by using object templates and additive independently and identically distributed Gaussian-Laplacian noise assumption, so that the tracker is well adopted to some complex scenes and is able to predict the accurate state of the object in every frame. Extensive experiments are carried out to validate that the online learning scheme improves the representation accuracy by exploring some specific properties of the target in every task, and the tracker maintains the best ability to achieve favorable performance. The experimental results show that the proposed algorithm performs favorably against several state-of-the-art tracking algorithms. 2015 Vol. 28 (3): 223-230 [Abstract] ( 467 ) [HTML 1KB] [ PDF 1604KB] ( 732 ) Researches and Applications 231 Structure Analysis for Human Models Based on Surface and Spatial Features HAN Li, LI Lin, XU Jian-Guo, TANG Di Structural feature of a 3D model has strong stability in affine transformation, and it effectively avoids the ambiguity of shape understanding. A method for 3D human structure analysis is developed by combining surface geometric feature with spatial feature. Aiming at the limitation of geodesic distance in revealing local part details of 3D human models, an approach is introduced for interior volume computation, and then it implements the structure detection and segmentation with the help of the fusion of surface shape feature and spatial structure feature. The proposed method not only improves the stability and applicability of shape analysis, but also enhances the accuracy of skeletal descriptor for 3D human model, and it can be further applied to the model identification and retrieval. 2015 Vol. 28 (3): 231-238 [Abstract] ( 480 ) [HTML 1KB] [ PDF 799KB] ( 760 ) 239 Selection of Skyline Representative Point YANG Li-Long, DONG Yi-Hong, HE Xian-Mang, QIAN Jiang-Bo Skyline query plays an important role in multi-decision and data mining. However, with the growth of data dimension, Skyline set becomes very large. Skyline representative point query is studied to overcome this shortcoming. A new evaluation function is proposed to improve the score-computing of Skyline points so as to select k representative Skyline points. A dynamic programming based algorithm(DPBA) in two-dimensional space is presented. The Eulerian distance between representative point and non-representative point is determined by the cover circle. k representative points are got by computing the evaluation function iteratively. In high-dimensional space, an approximate solution based on aR-tree index is proposed to solve the NP-hard problem. The index tree is traversed to judge whether it is dominated by the candidate Skyline sets. If it is dominated, it should be pruned to reduce the computation cost. The experiments of synthetic and real data show that the proposed algorithms are effective and efficient. 2015 Vol. 28 (3): 239-246 [Abstract] ( 442 ) [HTML 1KB] [ PDF 470KB] ( 604 ) 247 Unsupervised Feature Selection Based on Locality Preserving Projection and Sparse Representation JIAN Cai-Ren, CHEN Xiao-Yun Traditional filter-based feature selection methods calculate some scores of each feature independently to select features in a statistical or geometric perspective only, however, they ignore the correlation of different features. To solve this problem, an unsupervised feature selection method based on locality preserving projection and sparse representation is proposed. The nonnegativity and sparsity of feature weights are limited to select features in the proposed method. The experimental results on 4 gene expression datasets and 2 image datasets show that the method is effective. 2015 Vol. 28 (3): 247-252 [Abstract] ( 721 ) [HTML 1KB] [ PDF 654KB] ( 873 ) 253 A Recognition Approach for Cigarette Smoke Based on MI-Simba HU Chun-Hai, WANG Xiao-Jing, LIU Bin, SU Xiang-Yu, GUO Shi-Liang To overcome the uncertainty of smoke characteristics caused by the environment background, inhibit the redundancy between video smoke features, and improve the recognition rate simultaneously, a MI-Simba algorithm combining mutual information and simbafor recognizing cigarette smoke in indoor videos is proposed. Firstly, the statistic feature, color layout feature and dynamic feature of cigarette smoke are obtained by the method of video feature extraction, and then the initial feature vector is established. Secondly, the feature vector is updated automatically by MI-Simba, and the optimal feature combination in this environment is established. Then a transductive support vector machine(TSVM) is used for classification and recognition. Finally, the recognition rate and sensitivity are computed on the self-built video sequence database by 5-fold cross validation. The experimental results demonstrate the validity and superiority of the proposed algorithm compared with other algorithms. 2015 Vol. 28 (3): 253-259 [Abstract] ( 640 ) [HTML 1KB] [ PDF 888KB] ( 680 ) 260 An Improved Algorithm of Principal Line Extraction and RegistrationBased on MFRAT and ICP XIANG Bei-Hai, YU Zhao-Xian, QU Han-Bing To improve the recognition accuracy of palmprint images with some rotations and translations, the improved iterative closest point(ICP) algorithm based on direction and nearest neighbor(DNN) is proposed to make registration for principal-line feature images, and the strategy combining coarse registration and fine registration is taken in the process of principal-lines feature image registration. The experimental results show that the proposed method has high computational efficiency and good ability to resist rotation and translation. 2015 Vol. 28 (3): 260-265 [Abstract] ( 641 ) [HTML 1KB] [ PDF 565KB] ( 889 ) 266 Log-Gabor Feature-Based Nonlocal Means Denoising Algorithm and Its Acceleration Scheme ZHANG Song, JING Hua-Jiong The nonlocal means (NLM) is a spatial domain image denoising method, and it exploits long range similarities between pixels of natural images. Notably, the similarity between true pixel values in original NLM is estimated based on patch information of noise-corrupted input image. In this paper, the pixel similarities in NLM are estimated based on Log-Gabor features to achieve good denoising results. Moreover, the mixed similarity combining the Log-Gabor features with intensity information is exploited to get better adaptivity to local image characteristics andfurther improve the denoising quality. In addition, the random projection-based NLM speed-up method is studied based on Johnson-Lindenstrauss lemma. Extensive tests including the running time comparison before and after dimensionality reduction, the impact of types of projection matrices and the extent of dimensionality reduction on final denoising performances are carried out. The experimental results confirm the effectiveness of the proposed acceleration scheme. 2015 Vol. 28 (3): 266-274 [Abstract] ( 403 ) [HTML 1KB] [ PDF 656KB] ( 550 ) 275 Image Saliency Detection Based on Global and Local Information Fusion BAO Lei, LU Jian-Jiang, LI Yang, SHI Yan-Wei Visual Attention System is an important part of computer vision receiving more and more attention. In this paper, an image saliency detection model is presented based on global and local information fusion. The model firstly makes discrete shearlet decomposition on input image to obtain shearlet and scaling coefficients. As the shearlet coefficients contain most details of an image, a feature map is reconstructed on each decomposition level by performing inverse shearlet transform on these coefficients. Based on the feature maps, global and local contrasts are derived. On one hand, feature vectors are obtained by using all the feature maps to describe the detected image, and the global probability density distribution is calculated to obtain the global saliency value. After that, a global saliency map is obtained. On the other hand, the local entropy is calculated to measure the geometric distribution complexity of local areas on each feature map. After the local saliency value is obtained for every decomposition level, the local saliency map is built. By properly fusing global and local saliency maps, the total saliency map is obtained. The experimental results show that the proposed saliency detection model performs better than current models do. 2015 Vol. 28 (3): 275-281 [Abstract] ( 548 ) [HTML 1KB] [ PDF 696KB] ( 1130 ) 282 Multi-objective Optimization Control of Smith-Predictor Parameters in Superheated Steam Temperature System LIU Chang-Liang, MA Zeng-Hui The impact of Smith-predictor parameters on the control system performance is analyzed detailedly, and a multi-objective optimization control scheme of Smith-predictor parameters is proposed. The performance of the control system is improved by the model mismatch. Due to high order,large inertia and strong nonlinearity of the superheated steam temperature plant, a Smith-predictor parameter multi-objective self-tuning optimization control system is designed based on cascade PID. The control scheme is applied to a 600MW supercritical boiler superheated steam temperature control. The simulation results show that the proposed approach has a good robustness and can effectively overcome the long dead time and nonlinearity of the system, and it has much better performance compared with cascade PID and normal Smith predictor. 2015 Vol. 28 (3): 282-288 [Abstract] ( 478 ) [HTML 1KB] [ PDF 748KB] ( 538 )

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