Multiply oriented and curved handwritten text line dataset (VML-MOC)

VML-MOC (Visual Media Lab - Multiply Oriented and Curved) [1] is a natural handwritten benchmark dataset for heavily skewed and curved text lines. These text lines were written as remarks on the page margins by different writers over the years. They appear at different locations within the orientations that range between 0o and 180o or as curvilinear forms. VML-MOC dataset document images purely contain binarized side notes. Hence, the researchers can focus only on text line extraction of multiply oriented and curved text lines, devoid of dealing with the challenges of page segmentation, heterogeneity of side text and main text areas and binarization defects. The dataset consists of 30 document images divided into train (20 pages) and test (10 pages) sets. The ground truth is provided in three forms: raw pixel labeling, DIVA pixel labeling and PAGE xml file. References [1] B.Kurar, Rafi Cohen, I. Rabaev, and J. El-Sana VML-MOC: Segmenting a multiply oriented and curved handwritten text lines dataset. In the 3rd International workshop on Arabic and derived Script Analysis and Recognition (ASAR), pp. 13 - 18, 2019. https://ieeexplore.ieee.org/document/8892847

VML-MOC（视觉媒体实验室——多方向弯曲文本数据集）[1]是一款面向严重倾斜与弯曲文本行的自然手写基准数据集。该数据集内的文本行均为多年来不同书写者在页面页边空白处留下的批注，其既可以呈现为朝向介于0°至180°之间的任意方向，也可表现为曲线形态，且在页面内的分布位置各不相同。VML-MOC数据集的文档图像仅包含二值化的侧边批注，因此研究人员可专注于多方向、弯曲文本行的提取任务，无需处理页面分割、侧边文本与主体文本区域异质性以及二值化缺陷等各类挑战。该数据集共包含30幅文档图像，划分为训练集（20页）与测试集（10页）两个子集。真值标注（ground truth）以三种形式提供：原始像素标注、DIVA像素标注以及PAGE XML文件。参考文献[1] B. Kurar、Rafi Cohen、I. Rabaev与J. El-Sana：《VML-MOC：多方向弯曲手写文本行分割数据集》，收录于第三届阿拉伯语及衍生文字分析与识别国际研讨会（ASAR）论文集，第13-18页，2019年。https://ieeexplore.ieee.org/document/8892847