摘要
本文采用基于递归算法的去除离散点法消除孤立噪声,选用扫描边界的方法分割字符,来研究验证码自动识别技术,选择和提取稳定而又便于表示的特征向量是本系统的核心之一。本文提出了简单的字符特征提取方法:采用网格灰度特征并对该特征进行线性鉴别分析(LDA,Linear discriminant analysis)变换,结合最小距离分类器完成字符识别过程,通过提高训练样本数,有效解决了形近字符识别率低的问题,取得了很好的识别效果。
This thesis researches on technique of validation code recognition. It adopts a denoising approach based on recursire algorithm and employs the segmentation method through scanning borderline between characters. To select and extract stable and simple feature is key to the validation code recognition system. Thus we propose a simple character feature extraction approach. This approach calculates only one grid feature and transforms the original feature by LDA algorithm. Then the feature set is put into the minimum - distance classifier for recognition. In order to improve the low recognition rate for similar characters, the number of training set is increased. It is proved that borer recognition performance is achieved.
出处
《广东广播电视大学学报》
2009年第1期102-108,共7页
Journal of Guangdong Radio & Television University
关键词
验证码识别
线性鉴别分析
网格特征
validation code recognition
linear diseriminant analysis
grid feature
