mlsanityData to 14 Oct 2026

HiREN: Towards Higher Supervision Quality for Better Scene Text Image Super-Resolution

arXiv 2307.16410 ↗First posted 31 Jul 2023cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Scene text image super-resolution (STISR) is an important pre-processing technique for text recognition from low-resolution scene images. Nowadays, various methods have been proposed to extract text-specific information from high-resolution (HR) images to supervise STISR model training. However, due to uncontrollable factors (e.g. shooting equipment, focus, and environment) in manually photographing HR images, the quality of HR images cannot be guaranteed, which unavoidably impacts STISR performance. Observing the quality issue of HR images, in this paper we propose a novel idea to boost STISR by first enhancing the quality of HR images and then using the enhanced HR images as supervision to do STISR. Concretely, we develop a new STISR framework, called High-Resolution ENhancement (HiREN) that consists of two branches and a quality estimation module. The first branch is developed to recover the low-resolution (LR) images, and the other is an HR quality enhancement branch aiming at generating high-quality (HQ) text images based on the HR images to provide more accurate supervision to the LR images. As the degradation from HQ to HR may be diverse, and there is no pixel-level supervision for HQ image generation, we design a kernel-guided enhancement network to handle various degradation, and exploit the feedback from a recognizer and text-level annotations as weak supervision signal to train the HR enhancement branch. Then, a quality estimation module is employed to evaluate the qualities of HQ images, which are used to suppress the erroneous supervision information by weighting the loss of each image. Extensive experiments on TextZoom show that HiREN can work well with most existing STISR methods and significantly boost their performances.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 46 of its 82 references that are in the map. First posted on arXiv 31 Jul 2023.

Builds on

46 of 82 references.

References · 46

  1. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  2. 2014Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modelingcited by 2,367
  3. 2016Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Networkcited by 1,908
  4. 2021SwinIR: Image Restoration Using Swin Transformercited by 1,887
  5. 2019Towards VQA Models That Can Readcited by 1,718
  6. 2015Spatial Transformer Networkscited by 1,429
  7. 2020DocVQA: A Dataset for VQA on Document Imagescited by 1,317
  8. 2018Image Super-Resolution Using Very Deep Residual Channel Attention Networkscited by 1,264
  9. 2014Image Super-Resolution Using Deep Convolutional Networkscited by 1,224
  10. 2021Uformer: A General U-Shaped Transformer for Image Restorationcited by 866
  11. 2013Network In Networkcited by 605
  12. 2015An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognitioncited by 520
34 more
  1. 2019Scene Text Visual Question Answeringcited by 331
  2. 2017Learning a Single Convolutional Super-Resolution Network for Multiple Degradationscited by 228
  3. 2019Blind Super-Resolution With Iterative Kernel Correctioncited by 203
  4. 2021Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognitioncited by 194
  5. 2018Deep Learning for Single Image Super-Resolution: A Brief Reviewcited by 158
  6. 2022Scene Text Recognition with Permuted Autoregressive Sequence Modelscited by 149
  7. 2018Unsupervised Image Super-Resolution using Cycle-in-Cycle Generative Adversarial Networkscited by 133
  8. 2020Towards Accurate Scene Text Recognition with Semantic Reasoning Networkscited by 131
  9. 2020SEED: Semantics Enhanced Encoder-Decoder Framework for Scene Text Recognitioncited by 119
  10. 2017Focusing Attention: Towards Accurate Text Recognition in Natural Imagescited by 110
  11. 2021From Two to One: A New Scene Text Recognizer with Visual Language Modeling Networkcited by 97
  12. 2022SVTR: Scene Text Recognition with a Single Visual Modelcited by 88
  13. 2021Vision Transformer for Fast and Efficient Scene Text Recognitioncited by 81
  14. 2018NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognitioncited by 73
  15. 2020Scene Text Image Super-Resolution in the Wildcited by 63
  16. 2021Benchmarking Chinese Text Recognition: Datasets, Baselines, and an Empirical Studycited by 53
  17. 2020GTC: Guided Training of CTC Towards Efficient and Accurate Scene Text Recognitioncited by 52
  18. 2018Channel-wise and Spatial Feature Modulation Network for Single Image Super-Resolutioncited by 42
  19. 2020On Vocabulary Reliance in Scene Text Recognitioncited by 39
  20. 2021Text Prior Guided Scene Text Image Super-resolutioncited by 37
  21. 2022A Text Attention Network for Spatial Deformation Robust Scene Text Image Super-resolutioncited by 35
  22. 2019A Multi-Object Rectified Attention Network for Scene Text Recognitioncited by 30
  23. 2020Text Recognition in the Wild: A Surveycited by 29
  24. 2021Text Gestalt: Stroke-Aware Scene Text Image Super-Resolutioncited by 28
  25. 2019TextSR: Content-Aware Text Super-Resolution Guided by Recognitioncited by 26
  26. 2022C3-STISR: Scene Text Image Super-resolution with Triple Cluescited by 26
  27. 2021A Two-Stage Attentive Network for Single Image Super-Resolutioncited by 20
  28. 2018Edit Probability for Scene Text Recognitioncited by 19
  29. 2019Multi-grained Attention Networks for Single Image Super-Resolutioncited by 10
  30. 2022Real-World Image Super-Resolution by Exclusionary Dual-Learningcited by 10
  31. 2022Criteria Comparative Learning for Real-scene Image Super-Resolutioncited by 8
  32. 2022Scene Text Image Super-Resolution via Content Perceptual Loss and Criss-Cross Transformer Blockscited by 8
  33. 2023Towards Robust Scene Text Image Super-resolution via Explicit Location Enhancementcited by 8
  34. 2018Binary Document Image Super Resolution for Improved Readability and OCR Performancecited by 7

Built on by

Not cited yet.