基于TensorFlow和CNN模型的验证码识别研究  

Research on Captcha Recognition Based on TensorFlow and CNN Model

作　　者：马凯 贺晓松 MA Kai;HE Xiaosong(Chongqing Institute of Engineering,Chongqing 400056,China)

机构地区：[1]重庆工程学院,重庆400056

出　　处：《现代信息科技》2024年第13期65-69,共5页Modern Information Technology

基　　金：重庆工程学院校内科研基金(2022xzcr02)。

摘　　要：针对传统机器学习中应用于多位字符验证码的分割识别方法具有整体准确率低、泛化能力不足的问题,提出一种高效通用的识别方法。设计基于CNN模型的端到端字符型验证码识别流程,使用TensorFlow框架实现流程的数据训练和效果验证。该方法可以高效地识别出字符型验证码,其平均准确率为95%以上,输入整张图片,直接输出整体识别结果,具有更强的通用性。使用CNN模型识别多位字符验证码相比于传统机器学习方法具有更高的准确率和通用性。Aiming at the problems of low overall accuracy and insufficient generalization ability of segmentation and recognition methods applied to multi character captcha in traditional Machine Learning,an efficient and universal recognition method is proposed.It designs an end-to-end character captcha recognition process based on CNN model,and uses TensorFlow framework to implement data training and effectiveness verification of the process.This method can efficiently recognize character captcha with an average accuracy of over 95%.By inputting the entire image and directly outputting the overall recognition result,it has stronger universality.It uses CNN models to recognize multi character captcha has higher accuracy and versatility compared to traditional Machine Learning methods.

关 键 词：验证码识别 TensorFlow CNN 端到端 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]