The content on this page is the work of Professor Henry Baird.
Be advised that Professor Baird is no longer on the active faculty at Lehigh.
This content continues to be available as a courtesy, but it may not be maintained or current.
Lehigh University
COLLEGE HOME | LEHIGH HOME | SEARCH



•  Home
•  Research
•  Courses
•  Prof'l Activities
•  Conferences
•  Publications
•  Recent Talks
•  Patents
•  Awards
•  Miscellaneous
•  Vita (PDF)


   


Henry S. Baird

Research on Human Interactive

Proofs and CAPTCHAs


--- A joint project with Avaya Labs Research ---

Networked computers are vulnerable to cyber attacks in which programs---'bots, spiders, scrapers, spammers, etc---pose as human users in order to abuse services offered for individual human use. These abuses have included defrauding financial payment systems, spamming, stealing personal information, skewing rankings and recommendations, and ticket-scalping---and new ones arise every month.

Efforts to defend against such attacks have, over the last six years, stimulated investigations into a broad family of security protocols designed to distinguish between human and machine users automatically over networks and via GUIs, called CAPTCHAs: Completely Automated Public Turing tests to tell Computers and Humans Apart. I have developed three generations of reading-based CAPTCHAs:

  • PessimalPrint with Richard Fateman and Allison Coates of UC Berkeley [CBF03] [PDF]; 
  • BaffleText with Monica Chew of UC Berkeley [BC03] [PDF]; and
  • ScatterType with Terry Riopka, Michael Moll, & Sui-Yu Wang of Lehigh Univ [BR05] [PDF] and [BMW05] [PDF].
In addition, Jon Bentley of Avaya Labs Research and I have investigated 'implicit CAPTCHAs' [BB05] [PDF].  Extensions to this collaboration are presently supported by a Sponsored Research Agreement with Avaya.

A somehat broader research area, 'human interactive proofs' (HIPs), may be defined as challenge/response protocols which allow a human to authenticate herself as a member of a given group. Prof. Dan Lopresti of Lehigh's CSE Dept has demonstrated a speech-based CAPTCHA using synthesized speech with confusing background noise [LKS02].

Prof. Lopresti and I co-organized the 2nd Int'l HIP Workshop held here at Lehigh University May 19-20, 2005.

Virtually all commercial uses of CAPTCHAs exploit the gap in reading ability between humans and machines when confronted with degraded images of text. AltaVista, Yahoo!, PayPal, Microsoft, TicketMaster, and MailBlocks are among at least three dozen companies presently employing CAPTCHAs. The arms race between defenders of Web services and attackers is heating up.  Many CAPTCHAs have been broken by computer-vision attacks.

In our Pattern Recognition research lab, Prof. Lopresti and I are investigating a new generation of CAPTCHAs with the following properties.

  1. Able to detect break-ins automatically and able to respond to a break-in immediately by dropping in a fresh CAPTCHA that will be, with high confidence, able to repel the attack; a key element of this task is to design families of CAPTCHAs with graded levels of difficulty, both for humans and for machines; our long experience with principles underlying two generations of successful CAPTCHAs have convinced us that an essential determinant of success is to generate CAPTCHAs in regimes that lie within statistically well-characterized parameter spaces, such as those describing physics-motivated image degradations; these parameter spaces will, we strongly believe, also support the design of tunable-difficulty CAPTCHAs.
  2. Accessible to disabled users, e.g. people with visual impairments, reading disorders, etc); a key element of this task is to incorporate into the design of CAPTCHAs relevant knowledge from the existing literature on psychophysical investigations into normal and impaired human reading, speech recognition, and other perceptual modalities; in particular, we will design CAPTCHA technology which complies with the Americans with Disabilities Act.
  3. Welcoming of diverse user populations, e.g. people with relatively low tolerance for difficulty, little familiarity with English, fear of technology, etc.; a key element of this task is to design CAPTCHAs which are self-explaining, reassuring, rewarding, perhaps amusing -- and so effectively motivating, so that inexperienced users quickly grasp their purpose, understand the broad societal benefits, and so become willing to cooperate in increased networked security by solving CAPTCHA challenges.
  4. Robust, since multi-modal, using either (or both) visual and auditory challenges; a key element of this task is to design CAPTCHAs which pair visual and auditory challenges as alternatives, or present them sequentially for improved confidence, or “most excitingly'' integrate them into challenges whose correct solution requires users to combine cues simultaneously from both modalities.

[BBW05]  H. S. Baird, Michael A. Moll, and Sui-Yu Wang, "ScatterType: a Legible but Hard-to-Segment CAPTCHA," Proc., IAPR 8th Int'l Conf. on Document Analysis and Recognition, Seoul, Korea, August 29 - September 1, 2005.

[BR05] H. S. Baird and T. Riopka, "ScatterType:  a Reading CAPTCHA Resistant to Segmentation Attack," Proc., SPIE/IS&T Conf. on Document Recognition and Retrieval XII (DR&R2005), San Jose, CA, January, 2005.

[BB05] H. S. Baird and J. L. Bentley, "Implicit CAPTCHAs," Proc., SPIE/IS&T Conf. on Document Recognition and Retrieval XII (DR&R2005), San Jose, CA, January, 2005.

[CBF03] A. L. Coates, H. S. Baird, R. J. Fateman, "PessimalPrint: A Reverse Turing Test," Int'l. J. on Document Analysis & Recognition, Vol. 5, pp. 158-163, 2003.

[BC03] H. S. Baird and M. Chew, "BaffleText: a Human Interactive Proof," Proc., IS&T/SPIE Document Recognition & Retrieval X Conf. (DR&R2003), Santa Clara, CA, January 23-24, 2003.

[LKS02] D. Lopresti, G. Kochanski, and C. Shih, "Human Interactive Proofs for spoken language interfaces," In Proc., 1st Workshop on Human Interactive Proofs, Palo Alto, CA, pp. 30-34, January 2002.

image


© 2003 P.C. Rossin College of Engineering & Applied Science
Computer Science & Engineering, Packard Laboratory, Lehigh University, Bethlehem PA 18015