-
Notifications
You must be signed in to change notification settings - Fork 0
Related Work
Marvin edited this page Nov 8, 2016
·
1 revision
Completely Automated Public Turing Test to Tell Computers and Humans Apart
—> Based on OCR
- 2 different OCR algorithms for digitising old books.
- Words, for which both OCR’s fail, are one component used by recaptcha
- Recaptcha’s challenges consist of one word that both OCR’s failed to resolve and one word that was successfully recognized
- Each unknown word is sent to multiple users and their respective answers are compared to each other
- When they all agree by a minimum of 3 humans these words are solved and are used as control words for future captchas
- In case the answers differ desperately, this words get discarded as unreadable
—> Based on AI problems
- AI problems that could be solved by a fraction of people are not feasible for use as a captcha system - they should be solvable within a short amount of time
- “We do not allow captchas to base their security in the secrecy of a database or a piece of code”
- TL;DR - It is basically introducing AI problems that fits to the captcha requirement (instantly solvable by humans - hard for bots), e.g. image recognition
—> Focus on usability aspect while designing CAPTCHAS
-
Three main types of CAPTCHA’s:
- Text-based schemes(rely on distorted text images)
- rely on distorted text images(still recognisable by humans but hard for OCR)
- Sound-based schemes
- speech recognition
- Image-based schemes
- image recognition tasks
- Text-based schemes(rely on distorted text images)
-
Text-based Captcha’s are the most popular ones
- Main usability criteria:
- Learnability: How easy is it for users to accomplish basic tasks the first time they encounter the design?
- Efficiency: Once users have learned the design, how quickly can they perform tasks?
- Memorability: When users return to the design after a period of not using it, how easily can they re-establish proficiency?
- Errors: How many errors do users make, how severe are these errors, and how easily can they recover from the errors?
- Satisfaction: How pleasant is it to use the design?
- Captcha usability Criteria(quantitatively)
- Accuracy: how accurately can a user pass a CAPTCHA challenge? For example, how many times she has to try in order to pass a test?
- Response time: how long does it take for a user to pass the test?
- Perceived difficulty/satisfaction of using a scheme. How difficult to use do people perceive a CAPTCHA is? Are users subjectively satisfied and would they be willing to use such a scheme?
- Captcha usability dimensions —> How to improve accuracy, response time or perceived difficulty/satisfaction?
- Distortion. This dimension examines the form of distortions employed by a CAPTCHA and their impact on usability.
- Content. This dimension examines contents embedded in CAPTCHA challenges (or tests) and their impact on usability. For example, how should the content be organised, and is the content appropriate?
- Presentation. This dimension examines the way that CAPTCHA challenges are presented and its impact on usability
| Category | Usability issue |
|---|---|
| Distortion | Distortion method and level, confusing characters, friendly to foreigners? |
| Content | Character set, string length(How long? predictable or not?), random string or dictionary word?, offensive word? |
| Presentation | Font type and size, image size, use of colour, integration with web pages |
- Contrary to the common belief, text-based CAPTCHAs can be difficult for foreigners.
- Whether the length of strings used in a scheme is predictable or not can have interesting implications for both its security and usability.
- The use of colour in a CAPTCHA can have an impact on its usability, security or both.
- Overall, the design of CAPTCHA is still an art, rather than a science. It requires considerable study to evolve the design of secure and usable CAPTCHAs into a science.