期刊名称：IOP Conference Series: Earth and Environmental Science
印刷版ISSN：1755-1307
电子版ISSN：1755-1315
出版年度：2019
卷号：252
期号：5
页码：1-6
DOI：10.1088/1755-1315/252/5/052008
出版社：IOP Publishing
摘要：In this paper, the CAPTCHA recognition is studied by using neural network algorithm. Because the traditional BP neural network itself has problems such as uncertain structural parameters, low convergence rate and easy to fall into local minimum, this paper identifies the words in the image according to the pixel value of the verification code. In this paper, the image is divided into individual letters, and then the identified letters are spelled into words. Experimental results show that The accuracy of individual letters was higher, but the accuracy of letters identified from each small image was much lower. The accuracy can be improved by adding dictionaries to the prediction function to find the most matching words.
