Proceedings: GI 2008

Order and entropy in picture passwords

Saranga Komanduri , Dugald Hutchings

Proceedings of Graphics Interface 2008: Windsor, Ontario, Canada, 28 - 30 May 2008, 115-122

DOI 10.20380/GI2008.15

  • Bibtex

    @inproceedings{Komanduri:2008:10.20380/GI2008.15,
    author = {Komanduri, Saranga and Hutchings, Dugald},
    title = {Order and entropy in picture passwords},
    booktitle = {Proceedings of Graphics Interface 2008},
    series = {GI 2008},
    year = {2008},
    issn = {0713-5424},
    isbn = {978-1-56881-423-0},
    location = {Windsor, Ontario, Canada},
    pages = {115--122},
    numpages = {8},
    doi = {10.20380/GI2008.15},
    publisher = {Canadian Human-Computer Communications Society},
    address = {Toronto, Ontario, Canada},
    }

Abstract

Previous efforts involving picture-based passwords have not focused on maintaining a measurably high level of entropy. Since password systems usually allow user selection of passwords, their true entropy remains unknown. A 23-participant study was performed in which picture and character-based passwords of equal strength were randomly assigned. Memorability was tested with up to one week between sessions. The study found that both character and picture passwords of very high entropy were easily forgotten. However, when password inputs were analyzed to determine the source of input errors, serial ordering was found to be the main cause of failure. This supports a hypothesis stating that picture-password systems which do not require ordered input may produce memorable, high-entropy passwords. Input analysis produced another interesting result, that incorrect inputs by users are often duplicated. This reduces the number of distinct guesses users can make when authentication systems lock out users after a number of failed logins. A protocol for ignoring duplicate inputs is presented here. A shoulder-surfing resistant input method was also evaluated, with six out of 15 users performing an insecure behavior.