Captcha Recognition using convolutional neural networks with low structural complexity
Haolin Yang · Journal of Physics Conference Series · 2020
Abstract CAPTCHAs are automated tests designed to distinguish between humans and computers. They could be easily solved by humans, but they become challenging for machines to solve, therefore preventing programs from abusing online services and occupying internet resources. Using Convolutional Neural Networks, the CAPTCHA tests could be solved automatically at high efficiency. Current approaches of high accuracy CAPTCHA recognition can be structurally complicated. As a result, our team explored a different approach to solve CAPTCHAs using a Convolutional Neural Network that is more efficient in terms of structural complexity and run time, with image processing being a possibility to enhance the accuracy. We also tested our networks on CAPTCHA datasets with character adhesion and background noise.