Cropped Cursive Video Text Dataset (CCVTD)_Dataset

The Cropped Cursive Video Text Dataset (CCVTD) is a custom dataset developed for research in Urdu text recognition from video images. It consists of approximately 4,000 cropped images of Urdu cursive text extracted from online video platforms such as YouTube, Facebook, and Dailymotion. Each image is paired with a corresponding transcription (label file) that provides the ground truth text for supervised learning.

The dataset captures real-world challenges of Urdu cursive script, including connected characters, varying fonts, low resolution, and complex backgrounds. CCVTD is primarily used for training and evaluating OCR and video text recognition models, particularly deep learning approaches. It plays an important role in supporting research for low-resource languages like Urdu and improving text recognition in real-world video scenarios.