Table 6 Comparison of some recent CAPTCHA attacks
From: Human-artificial intelligence approaches for secure analysis in CAPTCHA codes
CAPTCHA | Attack method | Success rate | Type |
|---|---|---|---|
Gimpy, EZ-Gimpy | Shape context matching [78] | 33%, 92% | Text-based |
Megaupload CAPTCHA | Segmentation [79] | 78% | |
ReCAPTCHA | Neural networks [80] | 99.8% | |
Teabag3D, 3DCAPTCHA, Super CAPTCHA | Pixel extraction [19] | 31%, 58%, 27% | |
HelloCAPTCHA | PDM (Pixel Delay Map)/CL (Catching Line) [81] | 16% - 100% | |
NuCaptcha | Box shape analysis & SIFT algorithm [82] | 90% | |
Asirra | SVM (support vector machine) [83] | 82.7% | Image-based |
HumanAuth | Side-channel attack [84] | 92% | |
Google image-based CAPTCHA | Deep learning/CNN [85] | 70.78% | |
Facebook image-based CAPTCHA | Deep learning/CNN [85] | 83.5% | |
reCAPTCHA V2 | Deep learning/CNN [6] | 79–88% | |
Facebook image CAPTCHA | Deep learning/CNN [6] | 86% | |
China Railway CAPTCHA | Deep learning/CNN [6] | 90% | |
Avatar CAPTCHA | CNN [19] | 99% | |
FR-CAPTCHA | SVM [86] | 23% | |
FaceDCAPTCHA | SVM [86] | 48% | |
Minteye CAPTCHA | Sobel operators [87] | 100% | |
Tencent CAPTCHA | Deep learning/CNN [6] | 100% | |
Capy CAPTCHA, KeyCAPTCHA, Garb CAPTCHA | JPEG image continuity measurement [88] | 65.1%, 20%, 98.1% | |
CAPTCHaStar | Max concentration [89] | 96% | |
Audio reCAPTCHA | SVM [64] | 45–58% | Audio-based |
eBay audio CAPTCHAs | DFT (Discrete Fourier Transform) and supervised learning algorithm [65] | 75% | |
Microsoft and Yahoo audio | Non-continuous speech [66] | 49%, 45% | |
Audio reCAPTCHA | HMMs (Hidden Markov Models) [90], free online speech-to-text services, and minimal phonetic mapping [91] | 52%, 85.15% | |
GeeTest, Netease CAPTCHA | Sigmoid function [6] | 96%, 98% | Cognitive-based |
No CAPTCHA reCAPTCHA | Reinforcement learning [92] | 96–97% |