arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2104.02874cs.CV
收藏

基于动态剩余特征融合的文档布局分析

Document Layout Analysis via Dynamic Residual Feature Fusion

East China Normal University(华东师范大学);Videt Tech Ltd.(维迪特科技有限公司)

Xingjiao Wu, Ziling Hu, Xiangcheng Du, Jing Yang, Liang He

摘要:

文档版面分析(DLA)的目的是将文档图像分割成不同的感兴趣区域,并了解每个区域的作用,在光学字符识别(OCR)系统和文档检索等领域有着广泛的应用。然而,由于训练数据非常有限,并且缺乏有效的模型,因此构建DLA系统是一个挑战。在本文中,我们提出了一个端到端的联合网络命名为动态残差融合网络(DRFN)的DLA任务。具体而言,我们设计了一个动态的残差特征融合模块,可以充分利用低维信息,保持高维类别信息。此外,为了解决由于数据不足而导致的模型过拟合问题,提出了在有限的训练数据下进行有效微调的动态选择机制。我们用两个具有挑战性的数据集进行了实验,并证明了所提出的模块的有效性。

英文摘要:

The document layout analysis (DLA) aims to split the document image into different interest regions and understand the role of each region, which has wide application such as optical character recognition (OCR) systems and document retrieval. However, it is a challenge to build a DLA system because the training data is very limited and lacks an efficient model. In this paper, we propose an end-to-end united network named Dynamic Residual Fusion Network (DRFN) for the DLA task. Specifically, we design a dynamic residual feature fusion module which can fully utilize low-dimensional information and maintain high-dimensional category information. Besides, to deal with the model overfitting problem that is caused by lacking enough data, we propose the dynamic select mechanism for efficient fine-tuning in limited train data. We experiment with two challenging datasets and demonstrate the effectiveness of the proposed module.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑