Cursive-Text: A Comprehensive Dataset for End-to-End Urdu Text Recognition in Natural Scene Images

We present a comprehensive dataset for Urdu text detection and recognition in natural scene images. To develop the dataset, more than 2500 natural scene images were captured with a digital camera and a built-in mobile phone camera. Three separate datasets for isolated Urdu character images, cropped word images and end-to-end text spotting are developed. The isolated Urdu character image dataset contains 19901 images, while the cropped word image dataset contains 14100 cropped words. A lexicon of more than 40K commonly used Urdu words is also created. The ground truths for each of the image in the isolated character, cropped word or text spotting datasets are provided separately. The proposed datasets can be used to perform Urdu text detection, recognition or end-to-end recognition in natural scenes. These datasets can also be helpful to develop Arabic and Persian natural scene text detection and recognition systems, as Urdu is a derived language of these scripts and has many similar letters. The datasets can also be helpful to develop multi-language translation systems.

本研究提出了一款面向自然场景图像中乌尔都语（Urdu）文本检测与识别的综合性数据集。为构建该数据集，研究团队使用数码相机与内置手机摄像头采集了超过2500张自然场景图像。此外还构建了三个独立数据集，分别对应孤立乌尔都语字符图像、裁切后的单词图像以及端到端（end-to-end）文本spotting（text spotting）任务。孤立乌尔都语字符图像数据集共包含19901张图像，而裁切单词图像数据集则涵盖14100个裁切单词样本。此外还构建了包含4万余个常用乌尔都语单词的词表。孤立字符、裁切单词以及端到端文本spotting数据集的每张图像均配有独立的标注真值（ground truth）。所提出的数据集可用于开展自然场景下的乌尔都语文本检测、识别或端到端识别任务。由于乌尔都语脱胎于阿拉伯语与波斯语书写体系，且二者存在大量相似字符，因此本数据集也可用于研发阿拉伯语与波斯语的自然场景文本检测与识别系统。此外，该数据集还可用于研发多语言翻译系统。