合成文本行图像数据集

本研究创建了一个名为‘合成文本行图像数据集’的大型数据集，包含52,094条文本行图像，用于训练和验证文档图像质量评估模型。该数据集通过模拟真实文档图像的多种属性，如字体、背景和模糊度，来生成具有质量标签的文本行图像。创建过程中，使用了随机选择的中英文文本和背景，以及高斯模糊和旋转技术来模拟图像的实际质量变化。该数据集主要应用于文档图像质量评估领域，旨在提高自动文本识别和分析的准确性。

This study developed a large-scale dataset titled "Synthetic Text Line Image Dataset", which comprises 52,094 text line images for training and validating document image quality assessment models. This dataset generates text line images annotated with quality labels by simulating multiple attributes of real document images, including fonts, backgrounds, and blurriness. During the dataset construction, randomly selected Chinese and English texts and backgrounds, alongside Gaussian blur and rotation techniques, were utilized to simulate realistic quality variations of the images. This dataset is primarily applied in the domain of document image quality assessment, aiming to improve the accuracy of automatic text recognition and analysis.