International Conference on Computer and Knowledge Engineering

Home / 13th International Conference on Computer and Knowledge Engineering

Farsi Text in Scene: A new dataset

Authors :

Ali Salmasi¹ Ehsanollah Kabir²

1- Tarbiat Modares University 2- Tarbiat Modares University

Keywords :

Detection،Farsi،FTS،Persian،Dataset،OCR،Scene text

Abstract :

Due to the recent advancements in computer vision and scene understanding, detection & recognition of text within scene images have attracted significant interest from both academia and industry. To develop a reliable and effective text recognition system, having access to a suitable dataset is essential. While many datasets containing English text in scene images are available, the lack of a comparable dataset for Farsi is evident. To lay the groundwork for the advancement of Farsi text detection and recognition systems in the future, a new dataset of Farsi text in scene images is introduced. Images are taken by a smartphone mainly from shop panels and advertising banners in the street. Printed texts on the images are in different fonts, sizes, and colors on diverse and complex backgrounds which give this dataset variety needed for building a robust text detection and recognition system. In total, 1182 images are collected and annotated. To validate our dataset, we trained a deep object detection model and achieved an mAP50 of 46.4% on the test set.