Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-Domain Document Layout Analysis Using Document Style Guide

Xingjiao Wu, Luwei Xiao, Xiangcheng Du, Yingbin Zheng, Xin Li, Tianlong Ma, Cheng Jin, Liang He

TL;DR

This work tackles cross-domain document layout analysis by proposing an unsupervised DL-GDD framework that synergizes a document layout generator, a document elements decorator, and a document style discriminator. The approach enables generating high-quality, target-like documents without annotations and uses contrastive learning to align styles across domains, thereby bridging the gap between training and target data. Experiments across PubLayNet, CS-150, DSSE-200, and CDSSE show that style-guided cross-domain data significantly improves layout segmentation and quality assessment, outperforming baselines in multiple settings. The method offers a practical, annotation-free path to robust DLA generalization and provides a foundation for quality-aware data synthesis in cross-domain document analysis.

Abstract

The document layout analysis (DLA) aims to decompose document images into high-level semantic areas (i.e., figures, tables, texts, and background). Creating a DLA framework with strong generalization capabilities is a challenge due to document objects are diversity in layout, size, aspect ratio, texture, etc. Many researchers devoted this challenge by synthesizing data to build large training sets. However, the synthetic training data has different styles and erratic quality. Besides, there is a large gap between the source data and the target data. In this paper, we propose an unsupervised cross-domain DLA framework based on document style guidance. We integrated the document quality assessment and the document cross-domain analysis into a unified framework. Our framework is composed of three components, Document Layout Generator (GLD), Document Elements Decorator(GED), and Document Style Discriminator(DSD). The GLD is used to document layout generates, the GED is used to document layout elements fill, and the DSD is used to document quality assessment and cross-domain guidance. First, we apply GLD to predict the positions of the generated document. Then, we design a novel algorithm based on aesthetic guidance to fill the document positions. Finally, we use contrastive learning to evaluate the quality assessment of the document. Besides, we design a new strategy to change the document quality assessment component into a document cross-domain style guide component. Our framework is an unsupervised document layout analysis framework. We have proved through numerous experiments that our proposed method has achieved remarkable performance.

Cross-Domain Document Layout Analysis Using Document Style Guide

TL;DR

This work tackles cross-domain document layout analysis by proposing an unsupervised DL-GDD framework that synergizes a document layout generator, a document elements decorator, and a document style discriminator. The approach enables generating high-quality, target-like documents without annotations and uses contrastive learning to align styles across domains, thereby bridging the gap between training and target data. Experiments across PubLayNet, CS-150, DSSE-200, and CDSSE show that style-guided cross-domain data significantly improves layout segmentation and quality assessment, outperforming baselines in multiple settings. The method offers a practical, annotation-free path to robust DLA generalization and provides a foundation for quality-aware data synthesis in cross-domain document analysis.

Abstract

The document layout analysis (DLA) aims to decompose document images into high-level semantic areas (i.e., figures, tables, texts, and background). Creating a DLA framework with strong generalization capabilities is a challenge due to document objects are diversity in layout, size, aspect ratio, texture, etc. Many researchers devoted this challenge by synthesizing data to build large training sets. However, the synthetic training data has different styles and erratic quality. Besides, there is a large gap between the source data and the target data. In this paper, we propose an unsupervised cross-domain DLA framework based on document style guidance. We integrated the document quality assessment and the document cross-domain analysis into a unified framework. Our framework is composed of three components, Document Layout Generator (GLD), Document Elements Decorator(GED), and Document Style Discriminator(DSD). The GLD is used to document layout generates, the GED is used to document layout elements fill, and the DSD is used to document quality assessment and cross-domain guidance. First, we apply GLD to predict the positions of the generated document. Then, we design a novel algorithm based on aesthetic guidance to fill the document positions. Finally, we use contrastive learning to evaluate the quality assessment of the document. Besides, we design a new strategy to change the document quality assessment component into a document cross-domain style guide component. Our framework is an unsupervised document layout analysis framework. We have proved through numerous experiments that our proposed method has achieved remarkable performance.
Paper Structure (13 sections, 5 equations, 7 figures, 3 tables)

This paper contains 13 sections, 5 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Approach
  4. Network Architecture
  5. Document Layout Generator
  6. Document Elements Decorator
  7. Document Style Discriminator
  8. Experiments
  9. Implementation Details
  10. Datasets
  11. Qualitative Results
  12. Ablation Study
  13. Conclusions

Figures (7)

  • Figure 1: The first row contents comes from different data sets and those of the second row are from the same data set (DSSE-200 yang2017learning).
  • Figure 2: Illustration of the DL-GDD.
  • Figure 3: An example of the dataset.
  • Figure 4: The CS-150 real documents and their corresponding segmentation. Top: original. Middle: ground-truth. Bottom: predictions. Segmentation label colors are: figure, table, and No-Image.
  • Figure 5: The DSSE-200 real documents and their corresponding segmentation. Top: original. Middle: ground-truth. Bottom: predictions. Segmentation label colors are: figure, table, text and background.
  • ...and 2 more figures