Thai–English Multiscript Text Image (TEMS) Dataset

The Thai-English Multiscript Text Image (TEMS) Dataset is introduced as a publicly available resource for multilingual scene text recognition and related computer vision research. The dataset comprises 5,000 cropped text images systematically extracted from 1,625 natural scene photographs captured using smartphone cameras. Text regions were collected from diverse real-world environments, including billboards, commercial storefronts, road signs, menus, packaging, and publication covers, representing realistic visual conditions commonly encountered in practical applications.

The dataset contains Thai text, English text, mixed Thai-English text, numerals, punctuation marks, and special symbols. In total, the TEMS Dataset includes 161 unique character classes, comprising Thai script characters (44 consonants, 20 vowels, 4 tone marks, 3 punctuation marks, and 10 Thai numerals), English uppercase and lowercase letters (52 characters), Arabic numerals (10 digits), and 18 special symbols. The cropped text images exhibit substantial variation in text length, font style, illumination conditions, background complexity, and image dimensions, ranging from 45 to 1,512 pixels in width and from 18 to 275 pixels in height. These characteristics provide realistic and challenging conditions for multilingual scene text analysis and recognition tasks.