ALTID : Arabic/Latin Text Images Database for recognition research

This paper presents a new public offline database for Arabic/Latin printed and Arabic/Latin handwriting text. The database was developed to be employed in performance evaluation, result comparison and development of new methods related to document analysis and recognition. It may be used for, script identification, font identification, writer identification and word segmentation. The printed text is scanned from 731 pages of Latin and Arabic printed documents with grayscale format and 300 dpi resolutions. After preforming a manual segmentation, we obtained 1845 Arabic text and 2328 Latin text images. The handwritten dataset includes 460 Arabic and 582 Latin text-blocks which are written by 17 individuals with different ages and educational levels. Each text image of our database is provided with a ground truth file.