Annals of Faculty of Computer and Information Sciences, Hosei University
Department of digital media <<Previous Next>>
HOME >> No.7 CONTENTS >> Toru WAKAHARA
Professor
Toru WAKAHARA
Refereed Publications
  1. Shinya Makino and Toru Wakahara, "Affine-Invariant Recognition of Face Images Using GAT Correlation," in Proceedings of International Workshop on Advanced Technology 2006 (IWAIT2006), Vol.1, January 2006, pp. 279-284.
    Abstract - This paper addresses a challenging problem of performing normalization and recognition of face images at one time. The key idea is use of GAT (Global Affine Transformation) correlation for determining optimal 2D affine parameters that normalize a given image to yield the maximum correlation value with a target image. In our proposed method an input image is assigned to the enrolled face image associated with the largest GAT correlation value between the two images. By using 300 faces × 8 images (4 frontal and 4 near-frontal images) extracted from the public HOIP face image database subject to no normalization we show that the proposed method achieves a very high recognition rate of 99.79% as compared to that of 98.46% obtained by the well-known eigenface method as applied only to manually normalized face images.
  2. Minoru Yokobayashi and Toru Wakahara, "Binarization and Recognition of Characters in Scene Images Using a Maximum Separability Axis in Color Space and GAT Correlation," in Proceedings of Meeting on Image Recognition and Understanding (MIRU2006), Vol.1, July 2006, pp. 1030-1035.
    Abstract - This paper proposes a new technique of binarization and recognition of characters in color subject to a variety of image degradations and complex backgrounds. The key ideas are twofold. One is binarization using an optimal projection axis in the RGB color space that maximizes class separability between character and background. The other is recognition by global affine transformation (GAT) correlation that yields an affine-invariant maximum correlation value between input and template images. We use a total of 698 character images in natural scenes extracted from the public ICDAR 2003 robust OCR dataset. In advance, we classify those images into seven groups according to the degree of image degradations and/or background complexity. On the other hand, we only prepare a total of 62 single-font templates for alphanumerics. Experimental results show an average recognition rate of 81.2%, ranging from 94.5% for clear images to 39.3% for seriously distorted images.
  3. Minoru Yokobayashi and Toru Wakahara, "Binarization and Recognition of Degraded Characters Using a Maximum Separability Axis in Color Space and GAT Correlation," in Proceedings of 18th International Conference on Pattern Recognition (ICPR2006), Vol.2, August 2006, pp. 885-888.
    Abstract - This paper proposes a new technique of binarization and recognition of characters in color with a wide variety of image degradations and complex backgrounds. The key ideas are twofold. One is to automatically select one axis in the RGB color space that maximizes the between-class separability by a suitably chosen threshold for segmentation of character and background or binarization. The other is affine-invariant or distortion-tolerant grayscale character recognition using global affine transformation (GAT) correlation that yields the maximum correlation value between input and template images. In experiments, we use a total of 698 test images extracted from the public ICDAR 2003 robust OCR dataset containing a variety of single-character images in natural scenes. In advance, we classify those images into seven groups according to the degree of image degradations and/or background complexity. On the other hand, we only prepare a single-font set of 62 alphanumerics for templates. Experimental results show an average recognition rate of 81.4%, ranging from 94.5% for clear images to 39.3% for seriously distorted images.
  4. Hanako Kohmura and Toru Wakahara, "Determining Optimal Filters for Binarization of Degraded Characters in Color Using Genetic Algorithms," in Proceedings of the 18th International Conference on Pattern Recognition (ICPR2006), Vol.3, August 2006, pp. 661-664.
    Abstract -This paper proposes a new binalization technique of characters in color using genetic algorithms (GA) to search for an optimal sequence of filters through a filter bank. The filter bank contains simple image processing filters as applied to one of the RGB color planes and logical/arithmetic operations between two color planes. First, we classify images of degraded characters extracted from the public ICDAR 2003 robust OCR dataset into several groups according to degradation categories. Then, in the learning stage, by selecting training samples from each degradation category we apply GA to the combinatorial optimization problem of determining a filter sequence that maximizes the average fitness value calculated between the filtered training samples and their respective target images ideally binarized by humans. Finally, in the testing stage, we apply the optimal filter sequence to binarization of remaining test samples. Experimental results show the promising ability of the proposed method against a variety of image degradation causes.
  5. Shinya Makino and Toru Wakahara, "Evaluation of GAT Correlation's Ability in Affine-Invariant Matching of Gray-Scale Face Images," in Proceedings of First Korea-Japan Joint Workshop on Pattern Recognition (KJPR2006), Vol.1, November 2006, pp. 164-169.
    Abstract - This paper addresses a challenging problem of affine-invariant matching of face images. The enhanced GAT correlation technique demonstrates a marked ability for matching a canonical face image against artificially affine-transformed face images subject to rotation within 45 degrees, scale change within 50 percent, and translation within 25 percent of the face extent. Also, in recognition experiments using 300 faces × 8 images (4 frontal and 4 near-frontal images) in the public HOIP face image database as test samples the proposed method has achieved a very high recognition rate of 99.75% by only using simple and rough moment normalization as preprocessing.
Other Publications
  1. Satoshi Otaka and Toru Wakahara, "Fingerprint Verification by Core-Point-Based Perturbation Method," IEICE Technical Report, PRMU2006-159, pp. 177-182, November 2006.

>PAGE TOP