摘要
提出采用多小波神经网络簇伸展轮廓识别手写体数字的方法。该方法的原理是:跟踪待识别数字的轮廓,对轮廓进行均衡化和重采样,使其具有平移不变性和缩放不变性;采用多小波神经网络簇对轮廓壳进行伸展得到数级多分辨率和其平均值;将这些壳系数输入前馈神经网络簇,以识别该手写体数字。研究结果表明,该方法可用于将轮廓壳进行多分辨率分解。
A novel handwritten numeral recognition approach was presented using multi-wavelet neural network clusters to expand contour shell. The contour of the numeral is traced, then the contour is normalized and resampled so that it is translation- invariant and scale-invariant. Multiwavelet ortho-normal shell expansion is performed on the contour to get several resolution levels and average value. The shell coefficients are used as features to input into a feed-forward neural network to recognize the handwritten numerals. The results show that the ortho-normal shell decomposition can decomposes a signal into multi-resolution levels without down-sampling.
出处
《中南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2006年第2期356-360,共5页
Journal of Central South University:Science and Technology
基金
湖南省教育厅自然科学研究项目(02C429)
关键词
多小波变换
神经网络簇
手写体数字识别
multi-wavelet transform
neural network clusters
handwritten numeral recognition
