Post-OCR Paragraph Recognition by Graph Convolutional Networks

TL;DR

This paper introduces a graph convolutional network approach for paragraph recognition in document images, leveraging layout features and beta-skeleton graphs to improve efficiency and generalization over traditional methods.

Contribution

The paper presents a novel GCN-based method for paragraph recognition that is significantly smaller and more adaptable than existing R-CNN based models, with effective layout-based features.

Findings

Comparable or better accuracy than R-CNN models on PubLayNet

Model size is 3-4 orders of magnitude smaller

Good generalization from synthetic to real-world data

Abstract

We propose a new approach for paragraph recognition in document images by spatial graph convolutional networks (GCN) applied on OCR text boxes. Two steps, namely line splitting and line clustering, are performed to extract paragraphs from the lines in OCR results. Each step uses a beta-skeleton graph constructed from bounding boxes, where the graph edges provide efficient support for graph convolution operations. With only pure layout input features, the GCN model size is 3~4 orders of magnitude smaller compared to R-CNN based models, while achieving comparable or better accuracies on PubLayNet and other datasets. Furthermore, the GCN models show good generalization from synthetic training data to real-world images, and good adaptivity for variable document styles.