A Feature Selection for the Recognition of Handwritten Characters based on Two-Dimensional Wavelet Packet

2차원 웨이브렛 패킷에 기반한 필기체 문자인식의 특징선택방법

  • 김민수 (전남대학교 BK21 사업단) ;
  • 백장선 (전남대학교 통계학과) ;
  • 이귀상 (전남대학교 전산학과 정보통신연구소) ;
  • 김수형 (전남대학교 컴퓨터정보학부)
  • Published : 2002.08.01

Abstract

We propose a new approach to the feature selection for the classification of handwritten characters using two-dimensional(2D) wavelet packet bases. To extract key features of an image data, for the dimension reduction Principal Component Analysis(PCA) has been most frequently used. However PCA relies on the eigenvalue system, it is not only sensitive to outliers and perturbations, but has a tendency to select only global features. Since the important features for the image data are often characterized by local information such as edges and spikes, PCA does not provide good solutions to such problems. Also solving an eigenvalue system usually requires high cost in its computation. In this paper, the original data is transformed with 2D wavelet packet bases and the best discriminant basis is searched, from which relevant features are selected. In contrast to PCA solutions, the fast selection of detailed features as well as global features is possible by virtue of the good properties of wavelets. Experiment results on the recognition rates of PCA and our approach are compared to show the performance of the proposed method.

본 논문에서는 문자인식의 특징선택방법으로 2차원 웨이브렛 패킷을 이용하는 새로운 방법을 제안한다. 영상자료의 특징들로부터 중심특징을 선택하기 위한 차원축소 기법으로 주성분분석 기법이 주로 사용된다. 하지만, 주성분분석 기법은 고유시스템에 의존하기 때문에, 이상치나 잡음 등에 민감할 뿐만 아니라, 전역적 특징만을 선택하는 경향이 있다. 때때로, 영상자료의 중요한 특징이 가장자리 부분이나 뽀족한 부분 같은 지역적 정보일 수 있다. 이러한 경우, 주성분분석 기법은 좋은 결과를 줄 수 없다. 또한 고유시스템은 많은 계산시간을 요구한다. 본 논문에서 원 자료는 2차원 웨이브렛 패킷기저에 의해 변환되고, 최적 판별 기저가 탐색된 후, 그것으로부터 적절한 특징이 선택된다. 주성분분석 기법과 비교하여, 제안된 방법은 웨이브렛의 좋은 특성에 의해 전역적 특징뿐만 아니라 지역적 특징의 선택이 빠른 계산시간으로 이루어진다. 제안된 방법의 성능을 보이기 위해 PCA와 제안된 방법의 인식률의 실험결과가 분석되었다.

Keywords

References

  1. Kim, M. S. and Baek, J. S., 'Feature Extraction and Statistical Pattern Recognition for Image Data using Wavelet Decomposition,' The Korean Communications in Statistics, Vol. 6, No. 3, pp. 831-841, 1999 Kim, M.S.;Baek, J.S.
  2. Hotelling, H., 'Analysis of a complex of statistical variables into principal components,' J. Educ. Psych, 24, 1933
  3. Karhunen, K., 'Uber Linearen Methoden in der Wahrscheinlichkeitsrechnng,' Ann. Acad. Sci. Fennicae, Ser. A 37, no. 1, 1947
  4. Watanabe, S., 'Karhunen-Loeve expansion and factor analysis: theoretical remarks and applications,' Trans. 4th Prague Conf. Inform. Theory, Statist. Decision Functions, Random Processes (Prague), Publishing House of the Czechoslovak Academy of Sciences, pp. 645-660, 1967
  5. Saito, N. and Coifman, R. R., 'Local discriminant bases,' Mathematical Imaging : Wavelet Applications in Signal and Image Processing, 1994
  6. Learned, R. E. and Wilsky, A. S., 'A Wavelet Packet Approach to Transient Signal Classification,' Applied and Computational Harmonic Analysis, Vol. 2, pp. 265-278, 1995 https://doi.org/10.1006/acha.1995.1019
  7. Wickerhauser, M. V., Adaptive Wavelet Analysis From Theory to Software. AK peters: Boston, 1994
  8. Daubechies, I., 'Orthonormal bases of compactly supported wavelets,' Communications in Pure and Applied Mathematics, 41, pp. 909-996, 1988 https://doi.org/10.1002/cpa.3160410705
  9. Coifman, R. R. and Saito, N., 'Constructions of local orthonormal bases for classification and regression,' in Comptes Rendus Acad. Sci, Paris, Serie I 319, no. 2, pp. 191-196, 1994
  10. Coifman, R. R. and Wickerhauser, M. V., 'Entropy-based algorithms best basis selection,' IEEE Trans. Inform. Theory 38, no. 2, pp. 713-719, 1992 https://doi.org/10.1109/18.119732
  11. Collineau, S., 'Some remarks about the scalograms of wavelet transform coefficients,' In Wavelets and Their Applications, Byrnes, J. S., Byrnes, J. L., Hargreaves, K. A., and Berry, K. Kluwer Academic Publications: Dordrecht, The Netherlands, 1994
  12. Arino, M. A. and Vidakovic, B., 'On wavelet scalograms and their applications in economic time series,' Discussion Paper 95-21, ISDS, Duke University, Durham, North Carolina, 1995
  13. Basseville, M., 'Distance measures for signal processing and pattern recognition,' Signal Processing, 18, no. 4, pp. 349-369, 1989 https://doi.org/10.1016/0165-1684(89)90079-0
  14. Yamada, H., Yamamoto, K. and Saito, T., 'A Linear Normalization for Hand -printed Kanji Character Recognition-Line Density Equalization,' Pattern Recognition, Vol. 23, No. 9, pp. 1023-1029, 1990 https://doi.org/10.1016/0031-3203(90)90110-7