摘要
针对粘连字符分割致使验证码字符识别效果不佳的问题,提出一种带权值的模板匹配和基于监督学习的模板权值调整相结合的字符识别方法。该方法利用模板的灰度和权值,在验证码图像上搜寻兴趣点,再根据兴趣点的匹配度和兴趣点之间的欧式距离过滤掉次佳兴趣点,保留最佳兴趣点。利用基于目标像素个数期望的二值化阈值迭代优化,提高二值化质量和基于有监督的Hebb规则的模板权值学习提高识别率。通过与简单的模板匹配识别方法比较,实验结果表明,该方法对多网站验证码具有很好的识别率。
Traditional OCR methods are based on character segmentation,these methods are not suitable for CAPTCHA that has merged character.The paper puts out a non segmentation recognition method based on weighted template matching and supervised Hebb rule.This method employs gray level and weighted template,searches interesting point on the CAPTCHA image directly,filters out second-best interesting point by matching rate and Euclidean distance between each interesting point.Further more,a threshold optimization method for image thresholding and a weight adjusting method based on Hebb rule are included in the method.The result of the analysis and experiment proves that this method is more effective and feasible for CATPCHA recognition than simple template matching method.
出处
《计算机与现代化》
2010年第12期40-43,共4页
Computer and Modernization
关键词
权值模板
验证码
监督学习
相关匹配
weight template
CAPTCHA
supervised learning
correlation matching