Fujitsu R and nter Co.

Beijing, China

Fujitsu R and nter Co.

Beijing, China
Time filter
Source Type

Hou C.,Fujitsu Randnter Co. | Xia Y.,Fujitsu Randnter Co. | Xu Z.,Fujitsu Randnter Co. | Sun J.,Fujitsu Randnter Co.
Proceedings - International Conference on Pattern Recognition | Year: 2017

Classifier competence is critical important for dynamic classifier selection. This study proposes a semi-supervised learning algorithm to learn the competence of classifiers under the proposed optimization framework based on graph. First it constructs a graph based on the training data and some unlabeled data. Then it iteratively learns the competence of classifiers. The learned competence not just reflects the competitiveness of classifiers, but also varies smooth on the neighboring data. Experimental results on five different datasets show the dynamic classifier selection classification systems with the learned classifier competence perform better than the classification systems with local accuracy as the classifier competence. © 2016 IEEE.

He Y.,Fujitsu R and nter Co. | Sun J.,Fujitsu R and nter Co. | Naoi S.,Fujitsu R and nter Co. | Minagawa A.,Fujitsu Limited | Hotta Y.,Fujitsu Limited
Proceedings of SPIE - The International Society for Optical Engineering | Year: 2010

Quality of camera-based whiteboard images is highly related to the light environment and the writing effect of the content. Specular reflection and low contrast reduce the readability of captured whiteboard images frequently. A novel method is proposed to enhance camera-based whiteboard images in this paper. The images are enhanced by removing the highlight specular reflection to improve the visibility and emphasizing the content to improve the readability of the whiteboards. The method can be practically embedded in mobile devices with image capturing cameras. © 2009 Copyright SPIE - The International Society for Optical Engineering.

Huang K.,CAS Institute of Automation | Zheng D.,Fujitsu Randnter Co. | Sun J.,Fujitsu Randnter Co. | Hotta Y.,Fujitsu Limited | And 2 more authors.
Pattern Recognition Letters | Year: 2010

This paper provides a sparse learning algorithm for Support Vector Classification (SVC), called Sparse Support Vector Classification (SSVC), which leads to sparse solutions by automatically setting the irrelevant parameters exactly to zero. SSVC adopts the L0-norm regularization term and is trained by an iteratively reweighted learning algorithm. We show that the proposed novel approach contains a hierarchical-Bayes interpretation. Moreover, this model can build up close connections with some other sparse models. More specifically, one variation of the proposed method is equivalent to the zero-norm classifier proposed in (Weston et al., 2003); it is also an extended and more flexible framework in parallel with the Sparse Probit Classifier proposed by Figueiredo (2003). Theoretical justifications and experimental evaluations on two synthetic datasets and seven benchmark datasets show that SSVC offers competitive performance to SVC but needs significantly fewer Support Vectors. © 2010 Elsevier B.V. All rights reserved.

Tao Z.,Fujitsu R and nter Co. | Li L.,Fujitsu R and nter Co. | Liu L.,Fujitsu R and nter Co. | Yan W.,Fujitsu R and nter Co. | And 5 more authors.
IEEE Journal on Selected Topics in Quantum Electronics | Year: 2010

The Viterbi-and-Viterbi (V-V) algorithm is widely used to recover the carrier phase in optical digital coherent receivers. For simplicity, the basic V-V algorithm assumes constant carrier phase within the average duration. However, this basic assumption is probably violated by factors such as laser frequency offset and nonlinear XPM. In order to improve the basic V-V carrier phase recovery, five methods are introduced, verified, and analyzed. All these methods are compatible with parallel implementation that is mandatory for a realistic DSP circuit. The Q-improvement brought by each algorithm is analyzed together with the complexity of each. Among the five methods, the laser frequency offset compensation expands the tolerable frequency offset to ± 0.37 symbol rate, and the optimum weighted averaging in conjunction with normalization processing improves the Q -value by 2 dB under severe XPM condition. © 2010 IEEE.

Zhu Y.,Tianjin Normal University | Sun J.,Fujitsu Randnter Co. | Naoi S.,Fujitsu Randnter Co.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2012

In this paper, a natural scene character recognition method using convolutional neural network(CNN) and bimodal image enhancement is proposed. CNN based grayscale character recognizer has strong tolerance to degradations in natural scene images. Since character image is bimodal pattern image in essence, bimodal image enhancement is adopted to improve the performance of CNN classifier. Firstly, a maximum separability based color-to-gray method is used to strengthen the discriminative power in grayscale image space. Secondly, grayscale distribution normalization based on histogram alignment is performed. Through increasing the data consistency among grayscale training and test samples, it leads to a better CNN classifier. Thirdly, a shape holding grayscale character image normalization is adopted. Based on these measures, a high performance natural scene character recognizer is constructed. The recognition rate of 85.96% on ICDAR 2003 robust OCR dataset is higher than existing works, which verified the effectiveness of the proposed method. © 2012 Springer-Verlag Berlin Heidelberg.

Zhu Y.,Tianjin Normal University | Sun J.,Fujitsu R and nter Co. | Naoi S.,Fujitsu R and nter Co.
Proceedings of the International Conference on Document Analysis and Recognition, ICDAR | Year: 2013

This paper proposed a sub-structure learning based method for handwritten Chinese text recognition. In conventional methods, a standard character recognizer is trained on character classes only. Unreliable recognition results on character segments will decrease final recognition precision. By discovering stable sub-structure patterns from real character segment samples automatically, both character and sub-structure patterns are trained in character recognizer. The judgment reliability of segments being characters is significantly improved. Furthermore, to deal with millions of training segment samples, a two-stage clustering method is proposed for sub-structure learning. Experiment results on HIT-MW database show that the sub-structure learning based method improves performance significantly. The F1-measure evaluation of handwritten Chinese text recognition is improved by 8.84%. © 2013 IEEE.

Pan P.,Fujitsu R and nter Co. | Schonfeld D.,University of Illinois at Chicago
IEEE Signal Processing Letters | Year: 2011

In this letter, we extend the first-order Markov chain model commonly used in visual tracking and present a novel framework of visual tracking using high-order Monte Carlo Markov chain. By using graphical models to obtain conditional independence properties, we derive a general expression for the posterior density function of an m th-order hidden Markov model. We subsequently use Sequential Importance Sampling (SIS) to estimate the posterior density and obtain the high-order particle filtering algorithm for visual object tracking. Experimental results demonstrate that the performance of our proposed algorithm is superior to traditional first-order particle filtering (i.e., particle filtering derived based on first-order Markov chain). © 2006 IEEE.

Yin X.-C.,University of Science and Technology Beijing | Hao H.-W.,University of Science and Technology Beijing | Sun J.,Fujitsu R and nter Co. | Naoi S.,Fujitsu R and nter Co.
Proceedings of the International Conference on Document Analysis and Recognition, ICDAR | Year: 2011

Document images captured by a mobile phone camera often have perspective distortions. In this paper, fast and robust vanishing point detection methods for such perspective documents are presented. Most of previous methods are either slow or unstable. Based on robust detection of text baselines and character tilt orientations, our proposed technology is fast and robust with the following features: (1) quick detection of vanishing point candidates by clustering and voting on the Gaussian sphere space, and (2) precise and efficient detection of the final vanishing points using a hybrid approach, which combines the results from clustering and projection analysis. The rectified image acceptance rate for Mobile Cam-based documents, signboards and posters is more than 98% with an average speed of about 100ms. © 2011 IEEE.

Fu L.,Fujitsu R and nter Co. | Meng Y.,Fujitsu R and nter Co. | Xia Y.,Fujitsu R and nter Co. | Yu H.,Fujitsu R and nter Co.
Proceedings - 2nd International Conference on Information Technology and Computer Science, ITCS 2010 | Year: 2010

For web content extraction task, researchers have proposed many different methods, such as wrapper-based method, DOM tree rule-based method, machine learning-based method and so on. To some extent, all these methods ignore the layout information of the webpage, although the layout information such as the spatial and visual cues often plays a very important role in the process of locating the main content of the webpage when browsing. As a consequence, these methods often throw part of the main content away when extracting content from the webpage. In this paper, we present a method which combines webpage layout analysis with DOM tree rule-base method, it can make full use of the advantages of the two methods. It uses the layout information to guide the extraction work with a global view and can gain a better performance than the traditional methods. © 2010 IEEE.

He Z.,Fujitsu R and nter Co. | Meng Y.,Fujitsu R and nter Co. | Yu H.,Fujitsu R and nter Co.
Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference | Year: 2010

Hierarchical phrase-based models provide a powerful mechanism to capture non-local phrase reorderings for statistical machine translation (SMT). However, many phrase reorderings are arbitrary because the models are weak on determining phrase boundaries for patternmatching. This paper presents a novel approach to learn phrase boundaries directly from word-aligned corpus without using any syntactical information. We use phrase boundaries, which indicate the beginning/ ending of phrase reordering, as soft constraints for decoding. Experimental results and analysis show that the approach yields significant improvements over the baseline on large-scale Chineseto- English translation.

Loading Fujitsu R and nter Co. collaborators
Loading Fujitsu R and nter Co. collaborators