Annals of Faculty of Computer and Information Sciences, Hosei University
Department of computer science <<Previous Next>>
HOME >> No.2 CONTENTS >> Toru WAKAHARA
Professor
Toru WAKAHARA

Message
Having a lively intellectual curiosity in your study and research is most essential to taking a genuine delight in your academic life. In order to activate such curiosity, you have to think over what is an important problem worthy to be focused your energy on. In other words, finding a good problem is most valuable, and its solution is another thing.

>PAGE TOP

Publications(January 2001 - December 2001)
  1. Y. Kimura, T. Wakahara, and M. Sano, "Dictionary Learning by Adding Hyperspherical Templates," The Journal of the IIEEJ, Vol. 30, No. 1, pp. 11-20, January 2001 (in Japanese).
    Abstract - We present a new dictionary learning method in pattern recognition that generates a hyperspherical template with a small radius for each erroneous pattern embedded in the high dimensional feature space with decision boundaries of Voronoi regions. In recognition experiments on handwritten Kanji characters the proposed dictionary learning method shows a marked improvement in recognition accuracy with no over-learning effect.
  2. A. Suzuki, A. Shio, T. Wakahara, and S. Ohtsuka, "Using a Least Squares Filter to Improve the Binarization of Stripy Patterns," Proc. of International Workshop on Advanced Image Technology 2001 (IWAIT2001), pp. 45-48, Tajeon, February 2001.
    Abstract - This paper describes a new method that can improve the stability of binarization of degraded stripy patterns like wood rings or fingerprints while suppressing noise effectively. The proposed method utilizes the least squares filter via FFT as applied to each small block of the original image in order to enhance the local dominant frequency components of stripy patterns. Experimental results demonstrate the superiority of the proposed method over existing techniques.
  3. Y. Kimura, T. Wakahara, M. Sano, and A. Suzuki, "On-Line Recognition for Degraded Character Patterns Using Inter-Stroke Distance and Structural Information," The Journal of the IIEEJ, Vol. 30, No. 2, pp. 85-94, March 2001 (in Japanese).
    Abstract - We propose a new on-line character recognition method that combines the inter-stroke distance information with the relative structural information of constituent strokes by means of a weighted discriminant function. These two kinds of information play a complementary role in discriminating similar shaped but different characters. The proposed method realizes the reduction of the error rate by half in the extensive experiments of free-style Japanese handwritten character recognition.
  4. T. Wakahara, Y. Kimura, and A. Tomono, "Affine-Invariant Recognition of Gray-Scale Characters Using Global Affine Transformation Correlation," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 23, No. 4, pp. 384-395, April 2001.
    Abstract - This paper describes a new, promising technique of gray-scale character recognition that offers both noise tolerance and affine-invariance. The key ideas are the use of normalized cross-correlation as a matching measure and the application of global affine transformation (GAT) to the input image so as to achieve affine-invariant correlation with the target image. Recognition experiments show the high recognition accuracy against a wide range of rotation, scale change, and translation under random Gaussian noise as applied to gray-scale images of numerals.
  5. M. Mori and T. Wakahara, "Handwritten Kanji Character Recognition Using Relative Direction Contributivity," The Transactions of the IEICE, Vol. J84-D-II, No. 7, pp. 1360-1368, July 2001 (in Japanese).
    Abstract - In handwritten Kanji character recognition stroke directional features are effective. However, they are not robust against slanting, rotation, and the fluctuation of stroke direction. We propose new features that express the relative position and angle information based on directional features of adjacent strokes. Recognition experiments using the public database of ETL9B and artificially generated characters subject to heavy geometrical distortion show the substantial improvement in recognition rates.
  6. T. Wakahara, Y. Kimura, and M. Sano, "Handwritten Japanese Character Recognition Using Adaptive Normalization by Global Affine Transformation," Proc. of 6th International Conference on Document Analysis and Recognition (ICDAR'01), pp. 424-428, Seattle, September 2001.
    Abstract - This paper describes a new character recognition system with a category-dependent normalization technique that compensates for shape distortion between an input pattern and each reference pattern using global affine transformation (GAT). GAT adaptive normalization is applied to a set of candidate categories of the input pattern output by the basic OCR. Then, each adaptively normalized input pattern is fed again to the basic OCR. Recognition experiments using totally unconstrained handwritten characters demonstrate the substantial reduction of error rates.

>PAGE TOP