Fumitaka Kimura

Mie University, Tu, Mie, Japan

Are you Fumitaka Kimura?

Claim your profile

Publications (166)28.56 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: Extraction of facial feature points is essential task for many kinds of applications of face images. The feasibility of facial features extraction is determined by not only its accuracy but processing time. Some applications require real-time detection of facial features. This research aims to propose a method of facial features extraction by an accelerated implementation of circular Hough transform with gradients and appearance evaluation by histogram of gradient features. Experiment using FERET database shows that the proposed method successfully extracted eyes, nose and mouth for 98.44%, 99.50% and 98.79% of frontal face images in the dataset.
    No preview · Article · Dec 2015 · IEEJ Transactions on Electronics Information and Systems
  • K. Kuramoto · W. Ohyama · T. Wakabayashi · F. Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: Camera-Based Optical Character Recognition (CBOCR) has attracted interests of many researchers in both computer vision and document analysis research fields. A significant challenge in CBOCR is how we handle characters of those appearances are affected by three-dimensional (3D) rotation due to locational relationship between a printing plane and camera. Proper handling of these 3D rotated characters is expected to improve the performance of both detection and recognition of camera-captured characters. In this paper, we propose an efficient implementation of 3D rotation estimation for camera-captured characters. The proposed implementation requires small memory load and short computational time. We employ Linear Discriminant Function (LDF) instead of Modified Quadratic Discriminant Function (MQDF) for further memory reduction. The results of experimental evaluation using a large-scale alphanumeric character dataset showed that small number of dimensionality of original feature vector is sufficient for keeping accuracy of 3D rotation estimation and total amount of memory required for 3D rotation estimation is reduced from 141.0 MB to 6.6 MB.
    No preview · Article · Jul 2015
  • Mizuki Ito · Wataru Ohyama · Tetsushi Wakabayashi · Fumitaka Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: The performance of face recognition is easily affected by appearance variation by face rotation. The proposed method in this research recognizes who is a subject in the query image in which a face is captured from an arbitrary direction. The proposed method employs an auto-associative neural network for learning a manifold which represents principal variation of facial appearance in feature space due to face rotation. Our comparison where four conditions of selecting training samples for manifold learning were adopted implied that rotated third parson faces and its reference frontal face can be applicable for the manifold learning. The results in evaluation experiments with SCface database showed that the highest recognition accuracy at RANK10 is 77.5 %.
    No preview · Article · May 2015
  • H. Iwasa · W. Ohyama · T. Wakabayashi · F. Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: Extraction of facial feature points is essential task for many kinds of applications of face images. The feasibility of facial features extraction is determined by not only its accuracy but processing time. Some applications require real-time detection of facial features. This research aims to propose a method of facial features extraction by an accelerated implementation of circular Hough transform with gradients and appearance evaluation by histogram of gradient features. The acceleration implementation employs General Purpose computing on Graphics Processing Unit. Experiment using FERET database shows that the proposed method successfully extracted eyes and nose for 98.44% and 99.50% of frontal face images in the dataset. And 96.5% of computational time was reduced by accelerated implementation employing GPGPU.
    No preview · Article · May 2015
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a novel approach for offline Bangla (Bengali) handwritten word recognition by Hidden Markov Model (HMM). Due to the presence of complex features such as headline, vowels, modifiers, etc., character segmentation in Bangla script is not easy. Also, the position of vowels and compound characters make the segmentation task of words into characters very complex. To take care of these problems we propose a novel method considering a zone-wise break up of words and next perform HMM based recognition. In particular, the word image is segmented into 3 zones, upper, middle and lower, respectively. The components in middle zone are modeled using HMM. By this zone segmentation approach we reduce the number of distinct component classes compared to total number of classes in Bangla character set. Once the middle zone portion is recognized, HMM based forced alignment is applied in this zone to mark the boundaries of individual components. The segmentation paths are extended later to other zones. Next, the residue components, if any, in upper and lower zones in their respective boundary are combined to achieve the final word level recognition. We have performed a preliminary experiment on a dataset of 10,120 Bangla handwritten words and found that the proposed approach outperforms the custom way of HMM based recognition.
    No preview · Article · Dec 2014
  • [Show abstract] [Hide abstract]
    ABSTRACT: Accurate character recognition is still very important for camera based printed document analysis. Due to its inherent conceptual and technical simplicity, conventional recognition strategies relied on features extracted using square block zoning of a character image. In this paper, we propose an isotropic feature extraction method using regular hexagonal zoning and confirm its effectiveness by printed character recognition experiments. A 2% improvement in accuracy was achieved in experiments using gradient features. And the effectiveness of hexagonal zoning for recognition of high stroke count characters and low-resolution characters is confirmed by the experiments.
    No preview · Article · Dec 2014
  • Xi Luo · Wataru Ohyama · Tetsushi Wakabayashi · Fumitaka Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we present an effective way of combining character-based (N-gram) and word-based approaches for Chinese text classification. Uni-gram and bi-gram features are considered as the baseline model, which are then combined with word features of length greater than or equal to 3. A weight coefficient that can be used to give higher weights to word features is also introduced. We further employ a serial approach based on feature transformation and dimension reduction techniques. The results of McNemar's test indicate that the performance is significantly improved by our proposed method. © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.
    No preview · Article · Dec 2014 · IEEJ Transactions on Electrical and Electronic Engineering
  • Chen Chao · Wataru Ohyama · Tetsushi Wakabayashi · Fumitaka Kimura

    No preview · Article · Sep 2014
  • Sukalpa Chanda · Debleena Das · Umapada Pal · Fumitaka Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: Recognition of offline musical symbols can aid in automatic retrieval of a particular piece of musical notation from a digital repository. Though some work on on-line Musical symbol notations exists, little work has been done on off-line recognition of the symbols. This article proposes a system for offline isolated musical symbol recognition. Efficacy of a texture analysis based feature extraction method is compared with a structural shape descriptor based feature extraction method coupled with a Support Vector Machine (SVM) classifier. Later three different kinds of feature selection techniques were also analyzed to gauge the contribution of each feature in the overall classification process. We compared our results with an existing method and we noted the proposed system exhibited encouraging results and it is better than existing method. The proposed system even worked better when we used MQDF classifier in place of SVM. In a five-fold cross validation experimental framework, considering 3795 music symbols we achieved 97.50% and 98.05% accuracy from SVM and MQDF classifiers, respectively when chain-code histogram features are applied.
    No preview · Conference Paper · Sep 2014
  • Source
    Alireza Alaei · Umapada Pal · P. Nagabhushan · Fumitaka Kimura

    Full-text · Dataset · Apr 2014
  • Source
    Alireza Alaei · Umapada Pal · P. Nagabhushan · Fumitaka Kimura

    Full-text · Dataset · Apr 2014

  • No preview · Article · Jan 2014 · Pattern Recognition Letters
  • [Show abstract] [Hide abstract]
    ABSTRACT: Accurate off-line character recognition is still very important for camera based printed document analysis. Due to its inherent conceptual and technical simplicity, conventional recognition strategies relied on features extracted using square block zoning of a character image. In this paper, we propose an isotropic feature extraction method using regular hexagonal zoning and empirically confirm its effectiveness for printed and handwritten character recognition. We accomplished printed character recognition and handwritten character recognition experiments using large-scale evaluating datasets. The average accuracy was improved by 2 % in experiments using gradient features. And the effectiveness of hexagonal zoning for recognition of high stroke count characters and low-resolution characters is confirmed in both printed and handwritten character recognition by the experiments.
    No preview · Article · Jan 2014 · IEEJ Transactions on Electronics Information and Systems
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper addresses the problem of detecting characters in natural scene image. How to correctly discriminate character/non-character is also a very challenging problem. In this paper, we propose new character/non-character discrimination technique using the rotation angle of characters to improve character detection accuracy in natural scene image. In particular, we individually recognize characters and estimate the rotation angle of those characters by our previously reported method and use the rotation angle for character/non-character discrimination. As the result of the character recognition experiment evaluating 50 alphanumeric natural scene images, we have confirmed the accuracy improvement of precision and \(F\)-measure by 9.37 % and 4.73 % respectively when compared to the performance with previously reported paper.
    No preview · Chapter · Jan 2014
  • [Show abstract] [Hide abstract]
    ABSTRACT: Recently, expectations for camera-based document analysis and recognition have increased by improved performance of digital camera devices. In this paper, we propose a rotation angle estimation method using Gray-Scale Gradient Feature and Modified Quadratic Discriminant Function (MQDF). This method can recognize characters and estimate the rotation angle of those characters rapidly. As the result of the evaluation experiment using printed alphanumeric character, we have confirmed that the low dimensional feature vector is sufficient to estimate the rotation angle of characters. Also, we reduced the number of used eigenvectors of the covariance matrix to calculate the MQDF while keeping estimation accuracy.
    No preview · Article · Jan 2014 · IEEJ Transactions on Electronics Information and Systems
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper proposes a new signature verification technique called combined segmentation-verification based on off-line features and on-line features. We use three different off-line feature vectors extracted from full name Japanese signature image and from the sub-images of the first name and the last name. The Mahalanobis distance for each offline feature vector is calculated for signature verification. The on-line feature based technique employs dynamic programming (DP) matching technique for time series data of the signatures. The final decision (verification) is performed by SVM based on the three Mahalanobis distances and the dissimilarity of the DP matching. In the evaluation test the proposed technique achieved 97.22% verification accuracy with even FRR and FAR, which is 3.95% higher than the best accuracy obtained by the individual technique. This result shows that the proposed combined segmentation verification approach improves Japanese signature verification accuracy significantly.
    No preview · Conference Paper · Nov 2013
  • Xi Luo · Wataru Ohyama · Tetsushi Wakabayashi · Fumitaka Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we study on Chinese text classification using character-based approach (N-gram) and word-based approach and propose the use of uni-gram, bi-gram and word features of length greater than or equal to three. A weight coefficient which can be used to give higher weights to word features is also introduced. We further investigate a serial approach based on feature transformation and dimension reduction techniques to improve the performance. Experimental results show that our proposed approach is efficient and effective for improving the performance of Chinese text classification.
    No preview · Conference Paper · Jan 2013
  • [Show abstract] [Hide abstract]
    ABSTRACT: Although there are some reports on offline Tamil isolated handwritten character recognition, to our knowledge there is only two reports on Tamil off-line handwritten word recognition. Also no city name dataset is available for Tamil script. In this paper we present a Tamil offline city name dataset, we developed, and propose a scheme for recognition. Because of the different writing style of various individuals, some of the characters in a Tamil city name may touch and accurate segmentation of such touching into individual characters is a difficult task. Avoiding proper segmentation here, we consider a city name string as a word and the recognition problem is treated as lexicon driven word recognition. In the proposed method, binarized city names are pre-segmented into primitives (individual character or its parts). Primitive components of each city name are then merged into possible characters to get the best city name using dynamic programming. For merging, total likelihood of characters is used as the objective function and character likelihood is computed based on Modified Quadratic Discriminant Function (MQDF), where direction features are applied. A dataset of 265 Tamil city names is developed. and the database will be available freely to the researchers. From the experiment of the proposed scheme 96.89% city name accuracy is obtained from this dataset.
    No preview · Conference Paper · Jan 2013
  • Ryo Narita · Wataru Ohyama · Tetsushi Wakabayashi · Fumitaka Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we propose a new method for three dimensional rotation-free recognition of characters in scene. In the proposed method, we employ the Modified Quadratic Discriminant Function (MQDF) classifier trained with samples generated by three dimensional rotation process in a computer. We assume that when recognizing individual characters, considering three dimensional rotation can approximately handle the recognition of characters with perspective distortion. The results of the evaluation experiments using printed alphanumeric characters as an evaluation data set, consisting of approximately 600 samples per class for 62 character classes, show that the recognition rate is 99.34% for rotated characters while it is 99.59% for non rotated characters. We have empirically confirmed that the rotated characters given as the training data set do not negatively affect significantly to recognition of non rotated characters. Moreover, 813 characters extracted from 58 camera-captured scenes were correctly recognized and the feasibility of real world application of our method has been confirmed.
    No preview · Article · Jan 2013 · IEEJ Transactions on Electronics Information and Systems
  • Source
    Nilanjana Bhattacharya · Umapada Pal · Fumitaka Kimura
    [Show abstract] [Hide abstract]
    ABSTRACT: Recognition of Bangla compound characters has rarely got attention from researchers. This paper deals with segmentation and recognition of online handwritten Bangla cursive text containing basic and compound characters and all types of modifiers. Here, at first, we segment cursive words into primitives. Next primitives are recognized. A primitive may represent a character/compound character or a part of a character/compound character having meaningful structural information or a part incurred while joining two characters. We manually analyzed all the input texts written by different groups of people to create a ground truth set of distinct classes of primitives for result verification and we obtained 251 valid primitive classes. For automatic segmentation of text into primitives, we discovered some rules analyzing different joining patterns of Bangla characters. Applying these rules and using combination of online and offline information the segmentation technique was proposed. We achieved correct primitive segmentation rate of 97.89% from the 4984 online words. Directional features were used in SVM for recognition and we achieved average primitive recognition rate of 97.45%.
    Full-text · Conference Paper · Jan 2013

Publication Stats

2k Citations
28.56 Total Impact Points

Institutions

  • 1989-2015
    • Mie University
      • • Graduate School of Engineering
      • • Faculty of Engineering
      • • Department of Electrical and Electronic Engineering
      Tu, Mie, Japan
  • 2008
    • Tarleton State University
      SEP, Texas, United States
  • 1991-2007
    • University of Michigan-Dearborn
      • Department of Electrical and Computer Engineering
      Dearborn, Michigan, United States