期刊名称：International Journal of Computer Science and Network Security
印刷版ISSN：1738-7906
出版年度：2018
卷号：18
期号：2
页码：103-107
出版社：International Journal of Computer Science and Network Security
摘要：Over last decades or so, CAPTCHAs are used to differentiate between humans and bots. Although many alternatives of CAPTCHAs are introduced in recent years but still Text-based CAPTCHAs are most prevalent among all other alternatives on the internet. The traditional approach to recognize Text-based CAPTCHAs involves preprocessing, segmentation and finally recognition. This approach needs explicit segmentation of individual characters. Any weakness in prior steps leads to incorrect recognition. In this work, instead of using three-step approach, we have implemented a holistic approach to solve Text-based CAPTCHAs using a deep CNN. Significant improvements in the results have been achieved using our deep learning model. As deep leaning network need huge amount of data therefore we have synthetically generated two types of datasets i.e. easy and complex types of CAPTCHAs. Our model has shown an accuracy of 86.5 % and 83.3% on easy and complex datasets respectively.
