Explore chapters and articles related to this topic
Network-Based Interaction
Published in Julie A. Jacko, The Human–Computer Interaction Handbook, 2012
More radical are various forms of human computation (also known as crowd sourcing), where substantial tasks are achieved through the small actions of many people, often in the form of a game or puzzle. The most well-known of these is reCaptcha, which is used as a way to ensure users of a web page or service are human and not an automated agent (von Ahn et al. 2008). reCaptcha shows slightly distorted words, which the user needs to type in correctly in order to proceed in the site (see Figure 11.1). However, unlike many schemes where the words are algorithmically distorted from known text, in reCaptcha, the text displayed comes from documents where optical character recognition (OCR) has failed. One of the words is known, but the other is unknown, so that the user is effectively reading the unrecognized word and so slowly increasing the corpus of known text.
The Emerging Threat of Ai-driven Cyber Attacks: A Review
Published in Applied Artificial Intelligence, 2022
Blessing Guembe, Ambrose Azeta, Sanjay Misra, Victor Chukwudi Osamor, Luis Fernandez-Sanz, Vera Pospelova
Yu and Darling (2019) utilized an open-source Python Captcha package to boost recognition accuracy by combining TensorFlow object detection (TOD) and a speech segmentation method with CNN. The implemented model was able to determine which character was contained in a segmented sample. The result shows that the well-designed TOD+CNN model can crack open-source CAPTCHA libraries like Python Captcha and external captcha like the Delta40 benchmark. It has also been demonstrated that TOD+CNN can crack various types of CAPTCHAs, such as HashKiller13. Bursztein et al. (2014), developed a novel method for attacking captcha in a single step by combining segmentation and recognition problems using machine learning techniques. When both actions are done at the same time, the technique can take advantage of knowledge and context that would not be available if they were done separately. At the same time, it removes the need for any hand-crafted components, allowing this method to be applied to new Captcha schemes that the previous method could not. Without making any modifications to the algorithm or its settings, the authors were able to solve all of the real-world Captcha schemes they investigated exactly enough to consider the scheme insecure in reality, including Yahoo (5.33%) and ReCaptcha (33.34%). The success of this strategy against the Baidu (38.68%) and CNN (51.09%) schemes, both of which use occluding lines and character collapsing, implies that it can beat occluding lines in a broad sense. Noury and Rezaei (2020) proposed a vulnerability assessment Captcha solution based on deep learning. To explore the weaknesses and vulnerabilities of existing Captcha generation systems, the authors used a CNN model called DeepCAPTCHA. The numerical and alpha-numerical test datasets have cracking accuracy rates of 0.9894 and 0.983, respectively. That means more effort will be required to develop powerful Captchas that are resistant to AI-driven Captcha attack models.