Lightweight, highly accurate line and paragraph detection
6 hours ago
- #Document Layout Analysis
- #Computer Vision
- #Graph Convolutional Networks
- The paper presents a unified two-level clustering approach for detecting lines and paragraphs in documents.
- Text lines are clusters of text detection boxes, and paragraphs are clusters of lines, forming a two-level tree representing document layout.
- A graph convolutional network is used to predict relations between text detection boxes, enabling efficient and state-of-the-art paragraph detection.
- The method is tested on public benchmarks and real-world images, showing high efficiency and quality.