摘要
针对粘连字符验证码识别率低的问题,提出了一种基于模糊聚类和径向基神经网络的动态集成分类器。该分类器采用分割和识别反馈动态结合的思想,首先通过模糊聚类算法对字符进行特征提取,将其作为RBF神经网络的输入,然后网络依据识别置信度和字符特征隶属度进行特征节点的动态选择,最后通过实验进行了算法有效性和识别率的验证。与其他算法的对比实验进一步表明,该方法体现了整体优先,细节补偿的思想,能够充分利用训练样本集的信息,改进了低质量字符识别率不高的问题。
This paper puts forward a kind of integrated classifier based on fuzzy clusteringand the back propagation neural network, for the Captcha recognition with merged charac-ters. This classifier using of the dynamic feedback thought combination with segmentation and recognition. First, extracted feature of character by fuzzy clustering algorithm, and take it as the input of the RBF neural network. Then the network selects the node dynami-cally based on recognition confidence and membership degree of character features. Finally, Validation effectiveness and recognition rate through experiments. This classifier reflects the idea that considering the overall as priority and the details as compensation, making fulluse ving recognition rates of low quaity characters. of information of the training sample set, and impro-
出处
《东北电力大学学报》
2012年第4期40-43,共4页
Journal of Northeast Electric Power University
关键词
模糊聚类
RBF网络
识别置信度
验证码识别
Fuzzy clustering
RBF network
Recognition confidence
CAPTCHA recognition