COCO-Text-Patch

COCO-Text-Patch数据集由弗吉尼亚理工学院与州立大学等机构创建，包含约354,000个小图像，每个图像标记为‘文本’或‘非文本’。该数据集特别关注文本验证问题，这是文本检测和识别流程中的关键步骤。数据集的创建过程涉及从COCO-Text中提取32x32像素的小图像，并进行平衡处理，确保约半数图像包含文本，半数为背景。COCO-Text-Patch数据集适用于深度学习方法，旨在支持自动化文本分析系统的发展，解决日常场景中图像文本的检测和分析问题。

The COCO-Text-Patch dataset was developed by Virginia Polytechnic Institute and State University and other institutions. It contains approximately 354,000 small images, each annotated as either 'text' or 'non-text'. This dataset specifically targets the text verification task, which is a critical step in the text detection and recognition pipeline. The process of constructing this dataset entails extracting 32x32 pixel image patches from COCO-Text and conducting dataset balancing to ensure that roughly half of the samples include text, with the other half acting as background. The COCO-Text-Patch dataset is suitable for deep learning approaches, and is intended to support the development of automated text analysis systems to address text detection and analysis tasks for images captured in everyday real-world scenarios.