基于特征匹配与CNN的浮动验证码识别研究  

Research on the Identification of Floating CAPTCHA Based on Feature Matching and CNN

作　　者：胡蝶 侯俊[1] 何晨航 张磊[1] 陈伟朋 HU Die;HOU Jun;HE Chen-hang;ZHANG Lei;CHEN Wei-peng(School of Optical Electrical and Computer Engineering,Shanghai University of Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《软件导刊》2020年第5期37-41,共5页Software Guide

基　　金：上海自然科学基金项目(12ZR1420800)。

摘　　要：为有效识别浮动验证码,提出一种基于特征匹配与卷积神经网络的识别方法。首先使用特征匹配的方法得到匹配特征点,结合交叉匹配算法与K近邻匹配算法滤除错误匹配;然后对特征点进行聚类及投票分析,得到待识别字符区域,将其分割得到单个字符;最后在mnist手写数字数据集的基础上加入英文字符,构建卷积神经网络模型,将数据集送入模型进行训练。对10000张浮动验证码进行测试,结果表明,该方法对浮动验证码的识别准确率达95%,且构建的训练集具有可扩展性,可进一步应用到其它类型的字符识别中。In order to recognize floating captcha efficiently,a method based on feature matching and convolutional neural network is proposed.Firstly feature matching method is used to get the matched feature points.In order to filter out the wrong matches,cross matching algorithm and K-nearest neighbor matching algorithm are combined.Then,cluster and vote analysis are carried out on the feature points to get the character regions,which are segmented into single characters.Finally,English characters are added into mnist handwritten digital data set,the convolutional neural network model is constructed and the data set is fed into the model for training.In this paper,10,000 pieces of floating captcha are tested,and the results show that the recognition accuracy of floating captcha of the method reaches 95%,and the training set constructed is extensible,so the set can be further applied to other types of character recog⁃nition.

关 键 词：特征匹配 卷积神经网络 交叉匹配算法 K近邻匹配算法 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]