No Bot Expects the DeepCAPTCHA! Introducing Immutable Adversarial Examples, With Applications to CAPTCHA Generation
- Submitting institution
-
The University of Kent
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 9915
- Type
- D - Journal article
- DOI
-
10.1109/TIFS.2017.2718479
- Title of journal
- IEEE Transactions on Information Forensics and Security
- Article number
- -
- First page
- 2640
- Volume
- 12
- Issue
- 11
- ISSN
- 1556-6013
- Open access status
- Compliant
- Month of publication
- June
- Year of publication
- 2017
- URL
-
https://kar.kent.ac.uk/62081/
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
4
- Research group(s)
-
-
- Citation count
- 34
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- This paper is significant because, before this work, all CAPTCHAs were vulnerable to Deep Learning (DL) based attacks. We focused on an inherent limitation of DL algorithms, namely adversarial examples, to create DeepCAPTCHA, the first that is secure against DL-based attacks. Extensive testing against a large set of DL systems confirmed that DeepCAPTCHA is generalisable over different networks and offers significantly better security than existing CAPTCHAs. Moreover, a user study confirmed that DeepCAPTCHA offers better usability. This work opened a brand-new research avenue leading to numerous new CAPTCHA designs resistant to DL.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -