Restore Text First, Enhance Image Later: Two-Stage Scene Text Image Super-Resolution with Glyph Structure Guidance

2025-10-24

AI Summary

Existing generative super-resolution methods achieve strong performance on natural images but often distort character shapes when applied to text images, failing to simultaneously preserve text readability and visual fidelity. To address this, we propose TIGER, the first “text-first, then image” two-stage framework: Stage I employs a structure-aware network to reconstruct glyphs with high geometric accuracy; Stage II performs image super-resolution guided explicitly by the reconstructed glyphs. Our key innovation is establishing an explicit glyph-to-image guidance mechanism—breaking the longstanding trade-off between readability and perceptual quality. To support extreme-scale text super-resolution (×14.29), we introduce UltraZoom-ST, the first benchmark dataset specifically designed for scene text under severe degradation. Extensive experiments demonstrate that TIGER achieves state-of-the-art performance across multiple quantitative metrics, significantly improving character legibility and global visual consistency.

Technology Category

• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Current generative super-resolution methods show strong performance on natural images but distort text, creating a fundamental trade-off between image quality and textual readability. To address this, we introduce extbf{TIGER} ( extbf{T}ext- extbf{I}mage extbf{G}uided sup extbf{E}r- extbf{R}esolution), a novel two-stage framework that breaks this trade-off through a extit{"text-first, image-later"} paradigm. extbf{TIGER} explicitly decouples glyph restoration from image enhancement: it first reconstructs precise text structures and then uses them to guide subsequent full-image super-resolution. This glyph-to-image guidance ensures both high fidelity and visual consistency. To support comprehensive training and evaluation, we also contribute the extbf{UltraZoom-ST} (UltraZoom-Scene Text), the first scene text dataset with extreme zoom ( extbf{$ imes$14.29}). Extensive experiments show that extbf{TIGER} achieves extbf{state-of-the-art} performance, enhancing readability while preserving overall image quality.

Problem

• Restores distorted text in super-resolution images
• Decouples glyph reconstruction from image enhancement
• Breaks trade-off between image quality and readability

Innovation

• Two-stage framework decouples glyph restoration from enhancement
• Glyph structure guidance ensures high fidelity and consistency
• Novel dataset supports extreme zoom text super-resolution training