Image recognition problems in Google reCAPTCHA v2 are fixed using deep learning.

Abstract

The most widely used CAPTCHA solution nowadays is Google's reCAPTCHA v2, which provides primarily an
image-based CAPTCHA challenge. This research investigates the safety features of reCAPTCHA v2's image tests
and suggests a deep learning-based approach for quickly and easily bypassing them. Both a bespoke object
recognition deep learning model and Google's own Cloud Vision API are used to evaluate the suggested solution,
which uses human mouse mimicry to get around obstacles. Some countermeasures to improve reCAPTCHA v2's
security are also proposed in the article, along with some new avenues of attack.