A New Arabic (ahd/amsh) Handwritten Database

This paper introduces new database for Arabic handwritten words. The Arabic handwritten database (AHD/AMSH) represents a utility to facilitate the experiments of the character recognition algorithms. It contains three types of images: word, isolated character, and digit images. The AHD/AMSH can be used for baseline detection, characters segmentation, normalization, thinning, training and testing purposes. The stages of construction of the AHD/AMSH database were planned carefully to ensure its excellence. 150 words, 35 courtesy amount and 20 digits were used to fill the form which has been filled by 82 writers in 5 different age groups. The results were 12300 words, 29028 sub-words, 56170 characters, 2870 courtesy amounts, 820 Indian digits, and 820 Arabic digits. After dividing the database into two categories, training and testing, it has been tested manually and systematically. Keyword: Handwritten Arabic characters, Cursive writing, Database, Handwritten recognition