A Comprehensive Dataset of Complete Kurdish Handwritten Characters and Digits

The Kurdish language is written using two different styles of writing. Currently, it is written in Latin in Turkey and Syria, and Arabic in Iraq and Iran. Our dataset solely looks at Sorani (Central Kurdish) and mostly written with 34 characters and distributed over 4 different forms using a modified Arabic script. Since the Sorani script is written in a cursive way, the shape of each letter changes depending on its position within a word whether isolated, initial, medial, or final. Hopefully, our dataset has all the character forms.
This work introduces two comprehensive datasets: one encompassing the full set of Kurdish characters in the Central Kurdish script and another dedicated to handwritten digits. To collect handwritten samples systematically, each character set was printed on A4 sheets arranged in a grid of 12 × 14 cells, for each letter and digit, 84 copies of the sets have been prepared, guaranteeing a comprehensive and varied set of data and providing clear instructions for volunteers. Once the sheets were filled out, they underwent a thorough review to ensure accuracy. The forms were then scanned using high-quality equipment to capture the handwriting details effectively. Following the scanning process, Python programming was employed to segment, extract, and resize each character for further analysis. A total of 123,984 samples were collected, primarily from volunteer students and staff from various universities in Erbil, contributing to the diversity of the dataset.
The dataset includes handwritten images of Central Kurdish characters in their isolated, initial, medial, and final forms, along with digits from 0 to 9. Each character and digit have 1,008 samples. During the processing phase, every letter or digit was cropped into a separate image and saved in a dedicated folder, with each image assigned a unique ID. All images across the dataset were standardized in size.
Assigning Labels and Arrangements
The labeling process involved assigning three numbers to each image, separated by underscores. The first number represents the letter’s ID based on its alphabetical order. The second number corresponds to the form set, with 84 sets representing different groups of writers. The third number indicates the character’s or digit’s position in the form, ranging from 1 to 12, following a left-to-right order. For example, an image labeled "37_13_5.jpg" signifies that the character is the initial form of Pe (پـ), and the image occupies the fifth position in the form. Each letter was stored in a corresponding folder with 1,008 images for that specific letter or digit.

库尔德语目前存在两种书写体系。在土耳其与叙利亚地区，库尔德语采用拉丁字母书写；而在伊拉克与伊朗，则使用阿拉伯字母书写。本数据集仅聚焦索拉尼语（Sorani，中库尔德语），其书写基于改良阿拉伯字母，包含34个字符，并根据字符在单词中的位置分为四种不同形态。由于索拉尼文字以草书形式书写，每个字母的形态会根据其在单词中的位置（孤立、词首、词中、词尾）发生变化。本数据集旨在覆盖所有字符形态。

本工作推出了两套综合性数据集：一套涵盖中库尔德语书写体系的全部字符，另一套则专注于手写数字。为系统化采集手写样本，研究人员将每个字符集印刷在A4纸张上，以12×14的网格进行排版；针对每个字母与数字，共制备了84份重复样本集，以确保数据集的全面性与多样性，并为志愿者提供了清晰的书写指导。待填写完成后，工作人员会对所有纸张进行细致审核以保证数据准确性。随后使用高精度设备对填写完成的纸张进行扫描，以有效捕捉手写细节。扫描完成后，研究人员借助Python编程对每个字符进行分割、提取与尺寸调整，以供后续分析使用。本次采集共获得123984份样本，样本主要来自埃尔比勒多所高校的学生与教职工志愿者，这也提升了数据集的多样性。

本数据集包含中库尔德语字符的孤立、词首、词中、词尾四种形态的手写图像，以及0至9的手写数字图像。每个字符与数字均包含1008份样本。在数据处理阶段，每个字母或数字均被裁剪为独立图像并保存至对应专属文件夹，每张图像都会被分配唯一ID。数据集中所有图像均已完成尺寸标准化。

标签分配与编排规则
每张图像的标签由三个以下划线分隔的数字组成。第一个数字代表该字母按字母表顺序排列的ID。第二个数字对应书写者分组，共有84组不同的书写者。第三个数字表示该字符或数字在排版网格中的位置，取值范围为1至12，遵循从左至右的顺序。例如，标签为“37_13_5.jpg”的图像，表示该字符为字母پ（Pe）的词首形态，且该图像位于排版网格的第5个位置。每个字母均被存储在对应的专属文件夹中，每个文件夹包含对应字符或数字的1008份样本。