Organizations: The School of Marine Engineering, Dalian Maritime University, Dalian, China · The College of Artificial Intelligence, Dalian Maritime University, Dalian, China · The School of Information Science and Technology, Dalian Maritime University, Dalian, China · The Department of Information and Cyber Security, Shanghai Police College, Shanghai, China · The School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China
Ancient books preserve invaluable cultural heritage, yet their images often suffer from complex degradations caused by long-term ageing, environmental exposure, and physical damage. Existing ancient book image restoration benchmarks are limited in scale and rely on simplified degradation simulations, which cannot fully represent real-world degradation patterns and restrict the generalization ability of restoration models. To address these challenges, we introduce AncBench, a large-scale synthetic paired benchmark containing 100K aligned clean-degraded image pairs. AncBench models realistic composite degradations through a comprehensive degradation pipeline that considers material ageing, paper texture variation, blur, staining, water-induced geometric distortion, and local occlusion. Furthermore, we propose AncRestore, a degradation-aware visual-prior routing network for ancient book image restoration. AncRestore first learns degradation representations from degraded images and then dynamically selects and integrates suitable visual priors according to different degradation characteristics. Guided by these degradation cues, a dynamic routing mechanism combines expert-wise and region-wise routing strategies adaptively adjusts the contribution of these priors across different regions, enabling the model to recover both global appearance and local text details. Extensive experiments show that AncRestore effectively suppresses composite degradations while preserving text structures and enhancing restoration quality against other methods, highlighting its application potential for intelligent ancient book image restoration in cultural heritage preservation.
Figures & tables
Figure 1: Overview of the ancient book image restoration process. Intelligent techniques are used to repair damage, followed by text recognition preservation and sharing.
Figure 2: Construction and composition of the AncBench dataset. Clean ancient book images are synthetically degraded with chemical, physical, and biological damage using stochastic parameter; representative examples are shown on the right.
Figure 3: Overview of the proposed AncRestore. AncRestore combines degradation representation, HQS-based restoration backbone and dynamic expert routing to adaptively select and fuse multi-level visual priors for ancient-book image restoration.
Methods
Publication
Quality
Recognition
PSNR ↑
SSIM ↑
LPIPS ↓
MUSIQ ↑
CLIP-IQA ↑
Rec-Acc. ↑
Precision ↑
F1 score ↑
MPRNet
CVPR 2021
24.25
0.8979
0.1477
44.6083
0.3565
85.23%
74.60%
79.56%
Restormer
CVPR 2022
23.53
0.8897
0.1825
40.7427
0.3304
76.73%
75.08%
75.90%
PromptIR
NeurIPS 2023
23.63
0.8907
0.1764
41.0094
0.3370
76.82%
73.66%
75.21%
IR-SDE
ICML 2023
17.57
0.8130
0.2338
51.3281
0.3516
67.66%
70.91%
69.25%
SFHformer
ECCV 2024
24.26
0.8900
0.1765
41.9468
0.3651
81.59%
75.13%
78.23%
Table 1: Quantitative comparison on AncBench. Best and second-best results are highlighted in bold and underlined , respectively.
Figure 4: Qualitative comparison on AncBench. Red and green boxes show enlarged regions for detailed inspection. AncRestore more accurately reconstructs degraded character strokes while suppressing local artifacts and preserving the surrounding texture.
Methods
Publication
HDR28K
HanInpaint
Character-missing
Paper-damage
Ink-erosion
PSNR ↑
SSIM ↑
LPIPS ↓
MUSIQ ↑
PSNR ↑
SSIM ↑
LPIPS ↓
MUSIQ ↑
PSNR ↑
SSIM ↑
LPIPS ↓
MUSIQ ↑
PSNR ↑
SSIM ↑
LPIPS ↓
MUSIQ ↑
MPRNet
CVPR 2021
20.51
0.9012
0.1350
56.5054
21.85
0.9266
0.1093
55.9959
27.16
0.9574
0.0376
55.1148
28.94
0.9675
0.0480
59.5651
NAFNet
ECCV 2022
20.01
0.8976
0.0912
53.7241
15.43
0.8869
0.1514
54.5393
23.19
0.9358
0.0600
54.9008
31.72
0.9737
0.0363
60.4726
CharFormer
ICLR 2022
15.03
0.4247
0.5918
41.9007
16.90
0.8224
0.1764
59.0359
18.41
0.6253
0.3021
50.8800
24.66
0.9006
0.1090
58.5199
MambaIR
ECCV 2024
20.60
0.9016
0.1321
55.1465
21.24
0.9193
0.1154
55.2657
23.29
0.9354
0.0594
55.0047
28.71
0.9666
0.0447
59.1885
Table 2: Quantitative comparison on HDR28K and HanInpaint. Higher PSNR, SSIM, and MUSIQ values are better; lower LPIPS values are better. The best and second-best results are highlighted in bold and underline , respectively.
Figure 5: Qualitative comparison on HDR28K. Red and green boxes show enlarged regions for detailed inspection. AncRestore more accurately reconstructs degraded character strokes while suppressing local artifacts and preserving the surrounding texture.
Figure 6: Real-world restoration results, which trained on different synthetic ancient book image restoration datasets.
Variants
Components
Quality metrics
DER
DRB
PSNR ↑
LPIPS ↓
MUSIQ ↑
Baseline
✗
✗
25.56
0.1547
44.9086
Baseline+DRB
✗
✓
25.61
0.1553
43.9131
Baseline+DER
✓
✗
25.80
0.1558
44.1917
AncRestore
✓
✓
26.70
0.1346
45.2042
Table 3: Ablation studies on the effectiveness of the DER and DRB components in AncRestore.
Components
Setting
Quality metrics
PSNR ↑
LPIPS ↓
CLIP-IQA ↑
Routing
RR only
26.59
0.1485
0.3825
ER only
25.59
0.1568
0.3911
No. of P-Experts
1
26.01
0.1540
0.3874
2
25.92
0.1547
0.3915
3
26.55
0.1554
0.3922
Table 4: Ablations on the prior routing strategies and the numbers of private experts in DER.
Reliable document digitization in uncontrolled capture settings is challenging because real images exhibit multiple interacting degradations rather than a single isolated distortion. Documents thus captured are affected simultaneously by geometric distortions, like page warping, as well as photometric degradations such as non-uniform illumination, and blurring. However, most existing approaches address these factors independently and are evaluated on benchmarks containing only one distortion type, limiting their real-world applicability. We introduce GutenDoc, a large-scale dataset of high-resolution dense-text documents with physically grounded compound degradations. Using physics-based rendering, our dataset jointly models geometric warping and diverse photometric effects, enabling systematic evaluation under realistic capture conditions. We further propose a unified restoration framework that jointly corrects geometric and photometric distortions within a single-network and single-training setup, without the need for degradation-specific retraining or sequential inference passes. Extensive experiments show that our method remains competitive on established single-distortion benchmarks while substantially improving robustness under compound degradations, providing a practical solution for real-world document digitization.
Sujal Burad, Aakanksha, A. N. Rajagopalan +1
Indian Institute of Technology, Madras · Adobe, India
Ancient inscriptions frequently suffer missing or corrupted regions from fragmentation, erosion, or other damage, hindering reading, and analysis. We review prior image restoration methods and their applicability to inscription image recovery, then introduce MESA (Multi-Exemplar, Style-Aware) -an image-level restoration method that uses well-preserved exemplar inscriptions (from the same epigraphic monument, material, or similar letterforms) to guide reconstruction of damaged text. MESA encodes VGG19 convolutional features as Gram matrices to capture exemplar texture, style, and stroke structure; for each neural network layer it selects the exemplar minimizing Mean-Squared Displacement (MSD) to the damaged input. Layer-wise contribution weights are derived from Optical Character Recognition-estimated character widths in the exemplar set to bias filters toward scales matching letter geometry, and a training mask preserves intact regions so synthesis is restricted to damaged areas. We also summarize prior network architectures and exemplar and single-image synthesis, inpainting, and Generative Adversarial Network (GAN) approaches, highlighting limitations that MESA addresses. Comparative experiments demonstrate the advantages of MESA. Finally, we provide a practical roadmap for choosing restoration strategies given available exemplars and metadata.
Vasileios Toulatzis, Sofia Theodoridou, Ioannis Fudos
In Document Understanding, the challenge of reconstructing damaged, occluded, or incomplete text remains a critical yet unexplored problem. Subsequent document understanding tasks can benefit from a document reconstruction process. In response, this paper presents a novel unified pipeline combining state-of-the-art Optical Character Recognition (OCR), advanced image analysis, masked language modeling, and diffusion-based models to restore and reconstruct text while preserving visual integrity. We create a synthetic dataset of 30{,}078 degraded document images that simulates diverse document degradation scenarios, setting a benchmark for restoration tasks. Our pipeline detects and recognizes text, identifies degradation with an occlusion detector, and uses an inpainting model for semantically coherent reconstruction. A diffusion-based module seamlessly reintegrates text, matching font, size, and alignment. To evaluate restoration quality, we propose a Unified Context Similarity Metric (UCSM), incorporating edit, semantic, and length similarities with a contextual predictability measure that penalizes deviations when the correct text is contextually obvious. Our work advances document restoration, benefiting archival research and digital preservation while setting a new standard for text reconstruction. The OPRB dataset and code are available at Hugging Face and Github respectively.
Kunal Purkayastha, Ayan Banerjee, Josep Llados +1
Computer Vision Center · Indian Statistical Institute, Kolkata