DEVANAGARI CAPTCHA DATASET OF 1 Million Images : A challenge Test

CAPTCHA (Completely Automated Public Turing Tests to Tell Computers and Humans Apart). Only humans can successfully complete this test; current computer systems cannot. It is utilized in several applications for both human and machine identification. Text-based CAPTCHAs are the most typical type used on websites. Most of the letters in this protected CAPTCHA script are in English, it is challenging for rural residents who only speak their native tongues to pass the test. Devanagari characters have more complex characters than standard English characters and numeral-based CAPTCHAs, which makes machine recognition much more difficult. The majority of official websites in India only offer information in Devanagari. Unfortunately, websites do not use Devanagari CAPTCHAs.As a result, we have created a new text-based CAPTCHA in Devanagari script in this article. A computer/printed font and handwritten Devanagari character(34 each) and number(10 each) , in total 44+44 = 88 character images are used to design CAPTCHA. General CAPTCHA generation principles are used to add noise to the image using digital image processing techniques. Size of each CAPTCHA image is 250 X 90 pixels. 04 (Four) types of Character Sets are used – Printed Alphabet(34), Handwritten Alphabet(34), Printed Digit(10), and Handwritten Digit(10). Generated 11 Classes from these 04 combinations. The string length of the CAPTCHA image considered here is FIVE, SIX, and SEVEN ( 5, 6, 7). For each class – 03 (THREE) subclasses are created depending upon string length. In total there are 11 classes X 3 subclasses = 33 subclasses. So 33 types of CAPTCHA images were generated. For each class, 10,000 CAPTCHA images were created. For 11 Classes X 10,000 images , a Devanagari CAPTCHA Data set of 1,10,000 ( One Million Ten Thousand) images were created using Python. To make the CAPTCHA image less recognized or not easily broken. Passing a test with identifying Devanagari alphabets is difficult. It is beneficial to researchers who are investigating captcha recognition in this area. This dataset is helpful to researcher to design OCR for recognize Devanagari CAPTCHA and break it.CAPTCHA（全自动区分计算机与人类的图灵测试，Completely Automated Public Turing Tests to Tell Computers and Humans Apart）是一类仅人类可顺利完成、当前计算机系统无法攻克的测试。其被广泛应用于人机身份识别的多种场景中。基于文本的CAPTCHA是网站中最常见的类型。此类受保护的CAPTCHA脚本中的字母大多为英文字母，这使得仅掌握本土语言的乡村居民难以通过该测试。 天城文（Devanagari）字符相较于标准英文字母与基于数字的CAPTCHA更为复杂，这大幅提升了机器识别的难度。印度多数官方网站仅提供天城文界面，但遗憾的是，目前相关网站并未采用天城文CAPTCHA。为此，本文提出了一种全新的基于天城文脚本的文本型CAPTCHA。 本数据集采用计算机印刷字体与手写体天城文字符（各34种）及数字（各10种），总计88种字符图像（44种印刷体+44种手写体）用于CAPTCHA的构建。遵循通用CAPTCHA生成原则，通过数字图像处理技术为图像添加噪声。每张CAPTCHA图像的尺寸为250×90像素。本次实验采用四类字符集——印刷体字母（34种）、手写体字母（34种）、印刷体数字（10种）与手写体数字（10种），并基于这四类组合生成了11个类别。 本次研究考虑的CAPTCHA图像字符串长度为5、6、7三种。针对每一类，依据字符串长度划分出3个子类，总计11类×3子类=33个子类，即生成了33种不同类型的CAPTCHA图像。每类图像生成10000张，最终通过Python构建了包含110000（即十一万）张天城文CAPTCHA图像的数据集。 为降低CAPTCHA图像被识别或破解的难度，天城文字母识别测试本身具备较高难度。本数据集可为该领域内开展CAPTCHA识别研究的学者提供助力。 本数据集同样有助于研究者设计用于识别并破解天城文CAPTCHA的光学字符识别（OCR）系统。