Automated Korean-English Bilingual Highlighted Text Extraction Using HSV Segmentation

초록

This paper presents an automated system for extracting highlighted text from Korean-English documents using HSV-based color segmentation and OCR. By integrating HSV segmentation with linguistic correction methods, evaluation on a 600 -image dataset achieved mIoU of 0.8222 for highlight detection and 95.30% character-level accuracy (4.70% CER) for OCR. These results demonstrate that the combination of HSV segmentation and language-specific post-processing enables accurate and robust recovery of color-highlighted text for document digitization and analysis. © 2026 IEEE.

키워드