Analysis of Gabor parameters for handwritten numeral recognition by experimental design
Gabor filter-based features are useful for handprinted character recognition. One needs to optimize Gabor filter parameters because the performance of Gabor features depends strongly on Gabor filter parameters. One way to find the optimal values of the parameters is to analyze statistically the influence of the parameters on the error rate. In this paper, we discuss a statistical analysis of Gabor parameters for handwritten numeral recognition by experimental design. Our statistical analysis shows that optimal values of standard deviations σx and σy in Gabor filter are functions of the wavelength of the filter. In addition, it is shown that optimal values of σx and σy can be separately set on the condition that σx > σy.
- 1.Gabor, D.: Theory of communication. J. Inst. Elect. Engr. 93 (1946) 429–457Google Scholar
- 5.Hamamoto, Y., et al: Recognition of handprinted Chinese characters using Gabor features. Proc. 3rd Int. Conf. Document Analysis and Recognition (1995) 819–823Google Scholar
- 8.Yamada, K., Tsukumo, J.: Consideration on stability of Gabor feature extraction and character recognition application. Tech. Rep. IEICE Japan PRU92-112 (1993) 75–82 (in Japanese)Google Scholar
- 10.Montgomery, D.C.: Design and Analysis of Experiment, 4th Edition. John Wiley and Sons (1997)Google Scholar
- 11.Yamamoto, K.: Present state of recognition method on consideration of neighbor points and its ability in common database. MICE Trans. Information and Systems E79-D (1996) 417–422Google Scholar
- 12.Fukunaga, K.: Introduction to Statistical Pattern Recognition, 2nd Edition. Academic Press (1990)Google Scholar