Computers beat Humans at Single Character Recognition in Reading based Human Interaction Proofs (HIPs)

Human interaction proofs (HIPs) have become commonplace on the internet for protecting free online services from abuse by automated scripts/bots. They are challenges designed to be easily solved by humans, while remaining too hard for computers to solve. Reading based HIPs comprise a segmentation problem and one or more recognition problems. Recent studies have shown that computers are better at solving the recognition problem than the segmentation problem (Chellapilla and Simard, 2004; Chellapilla et al, 2005a). In this paper we compare human and computer single character recognition abilities through a sequence of human user studies and computer experiments using convolutional neural networks. In these experiments, we assume that segmentation has been solved and the approximate locations of individual HIP characters are known. Results show that computers are as good as or better than humans at single character recognition under all commonly used distortion and clutter scenarios used in todays HIPs.

[1]  Rachid Deriche,et al.  Fast algorithms for low-level vision , 1988, [1988 Proceedings] 9th International Conference on Pattern Recognition.

[2]  Henry S. Baird,et al.  Pessimal print: a reverse Turing test , 2001, Proceedings of Sixth International Conference on Document Analysis and Recognition.

[3]  Manuel Blum,et al.  Telling Humans and Computers Apart Automatically or How Lazy Cryptographers do AI , 2002 .

[4]  Henry S. Baird,et al.  BaffleText: a Human Interactive Proof , 2003, IS&T/SPIE Electronic Imaging.

[5]  Jitendra Malik,et al.  Recognizing objects in adversarial clutter: breaking a visual CAPTCHA , 2003, 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2003. Proceedings..

[6]  Patrice Y. Simard,et al.  Best practices for convolutional neural networks applied to visual document analysis , 2003, Seventh International Conference on Document Analysis and Recognition, 2003. Proceedings..

[7]  Patrice Y. Simard,et al.  Using Machine Learning to Break Visual Human Interaction Proofs (HIPs) , 2004, NIPS.

[8]  Joshua Goodman,et al.  Stopping outgoing spam , 2004, EC '04.

[9]  Mary Czerwinski,et al.  Designing human friendly human interaction proofs (HIPs) , 2005, CHI.

[10]  Mary Czerwinski,et al.  Building Segmentation Based Human-Friendly Human Interaction Proofs (HIPs) , 2005, HIP.