摘要
特征提取是手写体数字识别研究中的重要问题。有效特征是提高识别率和识别精度的关键。作者使用的主元分析法能压缩特征的维数 ,满足特征提取的完备性原则和正交性原则 ,提高分类器性能。将经过主元分析法压缩后的特征用BP神经网络进行识别仿真 。
Feature extraction is a principal problem for handwritten numerals recognition. A new method using K-L transform to extract effective features is proposed, which can compress feature dimension, fulfill the principle of integrity and irrelevance, and improve performance of classifier. Experiments with a BP neural network show the proposed approach is promising.
出处
《西华大学学报（自然科学版）》
CAS
2004年第S1期20-22,共3页
Journal of Xihua University:Natural Science Edition
关键词
特征提取
K-L变换
手写体数字识别
神经网络
feature extraction
K-L transform
handwritten numeral recognition
neural network
