Document Layout Analysis via Dynamic Residual Feature Fusion

TL;DR

This paper introduces DRFN, a novel end-to-end network for document layout analysis that effectively fuses features and adapts to limited training data, improving accuracy in understanding document regions.

Contribution

The paper presents a dynamic residual fusion module and a dynamic select mechanism, enhancing feature utilization and fine-tuning in data-scarce scenarios for document layout analysis.

Findings

Demonstrates improved performance on two challenging datasets.

Effectively handles limited training data with proposed mechanisms.

Validates the effectiveness of the dynamic residual fusion module.

Abstract

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.