ARASTI: A Database for Arabic Scene Text Recognition

Text in natural scenes provides many information for peoples and presents an essential tool to interact with their environment. Therefore, recognizing text existing in camera-captured images has become an important issue for many researches in the last decades. Currently, there isn't any available dataset of Arabic script text images in the wild. Since our aim is to help the research community in standardizing the evaluation of scene Arabic text recognition, we present in this paper a database of images of Arabic Scene Text, segmented scene Arabic words and segmented scene Arabic characters. We call this dataset ARASTI (ARAbic Scene Text Image). This database contains diverse natural scenes images captured at varying weather, lighting and perspective conditions. Moreover, characters and words are also segmented from the original images and stored individually. We obtain 1687 images, 1280 segmented scene Arabic words and 2093 scene Arabic character images. Compared to public datasets of scene text images in other languages like ICDAR03, Chars74K, etc., ARASTI contains a competitive number of images to these databases already published which proves that it can be used as a benchmark.