CAPTCHAs are widely used on the internet to determine whether a
user is a human, and text-based CAPTCHAs are mostly used. The study
on CAPTCHA recognition is meant for detecting the vulnerabilities in
their security for preventing any malicious intrusion in the network. In
this article, the Segmentation-based method and Segmentation-free
method are used for recognition. In segmentation-based technique, text
CAPTCHAs are segmented using contours and bounding-box method,
SIFT and KAZE features are extracted and Support Vector Machine
(SVM) and modified LeNet-5 model is used for recognition. In
Segmentation-free approach, we propose a customized Convolutional
Neural Network (CNN) for recognition. Really simple CAPTCHAs
dataset and Vulnerable CAPTCHAs dataset achieved a highest
recognition rate of 96.74% and 92.36% with pixel features using SVM.
Also, in modified LeNet-5 the highest recognition rates achieved for
these two datasets are 97.6% and 91.33% respectively. Using the
customized CNN without segmentation, these two datasets achieved
99.6% and 25.31% success rates. Also, some three different complex 5-
letter CAPTCHAs called ECE_1373_dataset, Captcha dataset and
Captcha_2000 are tested in this model and achieved 82.09%, 62.96%
and 61.33% accuracy rates.
S. Arivazhagan, M. Arun, F. Ruby Ranjitha Mary, B. Shamyughtha Bala
Mepco Schlenk Engineering College, India
CAPTCHA, SVM, CNN, LeNet-5
| January | February | March | April | May | June | July | August | September | October | November | December |
| 0 | 0 | 0 | 0 | 0 | 0 | 6 | 5 | 3 | 3 | 1 | 0 |
| Published By :
ICTACT
Published In :
ICTACT Journal on Data Science and Machine Learning
( Volume: 5 , Issue: 3 , Pages: 641 - 649 )
Date of Publication :
June 2024
Page Views :
124
Full Text Views :
18
|