In English

Solving CAPTCHA A study for testing the security of CAPTCHA

Anton Danielsson ; Filip Husnjak ; Linus Lindgren ; Isak Eriksson ; Patrik Wållgren ; Björn Larsson
Göteborg : Chalmers tekniska högskola, 2014. 39 s.
[Examensarbete för kandidatexamen]

The aim of this study is to create a computer application that can solve as many CAPTCHAs as possible. CAPTCHA is a security measure on the internet that stops automated programs from collection of sensitive information, cheating on online polls and spamming among other things. Given the amount of websites that depend on CAPTCHA as a security measure it is important to see whether or not it is possible to solve these under a rather short time frame. This project was done as an experimental study with a basis on previous attempts that have been published. The reports and ideas that other people have used are gone through and implemented together with new ideas. The main findings are that the noise protection, given to almost every CAPTCHA today, is not as safe as presumed as it is relatively easy for a computer to remove. The hardest part in a CAPTCHA for a computer to solve is the segmentation, which was not really solved in this project. Characters that are “fused” together are hard to segment from each other into single characters, which mean that the character recognition cannot recognise the characters. It was concluded that the CAPTCHAs are still a very safe way of protecting the websites from automated robots.

Nyckelord: CAPTCHA, Artificial Intelligence, Artificial Neural Network, Machine Learning, Optical Character Recognition, Image Segmentation, Image Distortion removal

Publikationen registrerades 2014-09-22.

CPL ID: 203116

Detta är en tjänst från Chalmers bibliotek