I am Robot: (Deep) Learning to Break Semantic Image CAPTCHAs

Sivakorn, Suphannee; Polakis, Iasonas; Keromytis, Angelos D.

doi:10.1109/eurosp.2016.37

Cited by 102 publications

(106 citation statements)

References 28 publications

Supporting

Mentioning

104

Contrasting

Unclassified

Order By: Relevance

“…As further evidence of this fact, our system shows promise of improvement over an approach for defeating Google's reCaptcha system published by Shet et al [3]. Their system achieves 70.78% accuracy [7]. Although direct comparisons are difficult due to differences in the underlying captcha implementations, the two captchas both require a user to select relevant images based on displayed text.…”

Section: Defeating the Captchamentioning

confidence: 63%

“…They state that only 6%, 27%, and 65% of people successfully pass the captcha on their first, second, and third attempts, respectively. provement over a recently released system by [7]. Their model solved about 71% of Google's image reCaptcha challenges [3], although direct comparisons of the two systems are difficult due to differences in the captcha implementations.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning to Associate Words and Images Using a Large-Scale Graph

Ya¹,

Sun²,

Helt³

et al. 2017

2017 14th Conference on Computer and Robot Vision (CRV)

View full text Add to dashboard Cite

We develop an approach for unsupervised learning of associations between co-occurring perceptual events using a large graph. We applied this approach to successfully solve the image captcha of China's railroad system. The approach is based on the principle of suspicious coincidence, originally proposed by Barlow [1], who argued that the brain builds a statistical model of the world by learning associations between events that repeatedly co-occur. In this particular problem, a user is presented with a deformed picture of a Chinese phrase and eight low-resolution images. They must quickly select the relevant images in order to purchase their train tickets. This problem presents several challenges: (1) the teaching labels for both the Chinese phrases and the images were not available for supervised learning, (2) no pre-trained deep convolutional neural networks are available for recognizing these Chinese phrases or the presented images, and (3) each captcha must be solved within a few seconds. We collected 2.6 million captchas, with 2.6 million deformed Chinese phrases and over 21 million images. From these data, we constructed an association graph, composed of over 6 million vertices, and linked these vertices based on co-occurrence information and feature similarity between pairs of images. We then trained a deep convolutional neural network to learn a projection of the Chinese phrases onto a 230dimensional latent space. Using label propagation, we computed the likelihood of each of the eight images conditioned on the latent space projection of the deformed phrase for each captcha. The resulting system solved captchas with 77% accuracy in 2 seconds on average. Our work, in answering this practical challenge, illustrates the power of this class of unsupervised association learning techniques, which may be related to the brain's general strategy for associating language stimuli with visual objects on the principle of suspicious coincidence.

show abstract

Section: Defeating the Captchamentioning

confidence: 63%

Section: Introductionmentioning

confidence: 99%

Learning to Associate Words and Images Using a Large-Scale Graph

Ya¹,

Sun²,

Helt³

et al. 2017

2017 14th Conference on Computer and Robot Vision (CRV)

View full text Add to dashboard Cite

show abstract

“…In order to try to find out the characters that compose the CAPTCHA is necessary to use some classification or recognize algorithm, generally for images or characters. There are some proposals in the literature about techniques to discover CAPTCHA using some classification algorithm, among them it is possible to mention: Support Vector Machine (SVM) [7], k nearest neighbor (k-NN) [2], deep learning [16] and many others [13]. Strategies to auto-fill CAPTCHAs are not something necessarily new.…”

Section: Introductionmentioning

confidence: 99%

“…Since its creation, the CAPTCHA methodology has been used for many website and a multitude of Internet services. Nowadays, CAPTCHA is a common strategy for websites to prevent attacks from anyone who try to have an automatic interaction with these websites, for example soft robots for automatic information extraction [16,10].…”

Section: Introductionmentioning

confidence: 99%

“…Golle's work [7] shows the use of SVM (support vector machines) to classify the images from Asirra CAPTCHA and he got a success rate of 82.7%. Sivakorn [16] used deep learning techniques to decipher images associates to the new model of CAPTCHA called reCAPTCHA getting good success rate 70.78% to the CAPTCHAs in the reCAPTCHA website and 83.5% in the CAPTCHAs from Facebook. Therefore, with the sophistication of the CAPTCHA methodology, new and more complex algorithms are proposed to automatization the systems access.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Simple Approach to Automatic Filling CAPTCHA using Pattern Recognition

Neto¹,

Batista²,

Ferreira³

2017

IJCA

View full text Add to dashboard Cite

This article shows an easy and simple approach to recognize characters in CAPTCHA images, where the k-NN (k nearest neighbor) algorithm is employed. This proposal to recognize characters in CAPTCHA images has the objective of autofill these components in order to support automation of access to systems. The main aim of this article is to show the steps involved in the proposed process about automatic filling CAPTCHAs since the image's handling until the classification of the characters through a simple and low-cost (implementation) technique of pattern recognition. Experimental results and an error distribution about the characters' classification are showed, where it is demonstrated the possibility of application in real cases of the proposal presented. General TermsCompletely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA)

show abstract

Battling bots: Experiences and strategies to mitigate fraudulent responses in online surveys

Goodrich

Fenton

Penn

et al. 2023

Applied Eco Perspectives Pol

View full text Add to dashboard Cite

Declining survey response rates have driven many researchers to seek out cost-effective methods of increasing participation, such as conducting surveys online, paying incentives, and using social media to engage hard-to-reach populations. Malicious actors can exploit the monetary incentives and anonymity of online surveys, threatening the integrity of survey data.We share two recent experiences conducting online surveys that were inundated with fraudulent responses. Our objective is to increase awareness of this emerging issue and offer guidance for others to mitigate the effects of fraudulent responders in their own research.

show abstract

I am Robot: (Deep) Learning to Break Semantic Image CAPTCHAs

Cited by 102 publications

References 28 publications

Learning to Associate Words and Images Using a Large-Scale Graph

Learning to Associate Words and Images Using a Large-Scale Graph

A Simple Approach to Automatic Filling CAPTCHA using Pattern Recognition

Battling bots: Experiences and strategies to mitigate fraudulent responses in online surveys

Contact Info

Product

Resources

About