Sketch-based manga retrieval using manga109 dataset

By Yusuke Matsui, Kota Ito, Yuji Aramaki, Azuma Fujimoto, Toru Ogawa, Toshihiko Yamasaki, Kiyoharu Aizawa. 2015

1,476 citations recorded by the source corpus.

Manga (Japanese comics) are popular worldwide. However, current e-manga archives offer very limited search support, i.e., keyword-based search by title or author. To make the manga search experience more intuitive, efficient, and enjoyable, we propose a manga-specific image retrieval system. The proposed system consists of efficient margin labeling, edge orientation histogram feature description with screen tone removal, and approximate nearest-neighbor search using product quantization. For querying, the system provides a sketch-based interface. Based on the interface, two interactive reranking schemes are presented: relevance feedback and query retouch. For evaluation, we built a novel dataset of manga images, Manga109, which consists of 109 comic books of 21,142 pages drawn by professional manga artists. To the best of our knowledge, Manga109 is currently the biggest dataset of manga images available for research. Experimental results showed that the proposed framework is efficient and scalable (70 ms from 21,142 pages using a single computer with 204 MB RAM).

Subfields: Computer Vision and Pattern Recognition, Information Retrieval & Search

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Information Retrieval & Search

Citations in the corpus

Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution (2017, neutral): “Among these datasets, SET5, SET14 and BSDS100 consist of natural scenes; URBAN100 contains challenging urban scenes images with details in different frequency bands; and MANGA109 is a dataset of Japanese manga.”
Image Super-Resolution Using Very Deep Residual Channel Attention Networks (2018, positive): “For Urban100 and Manga109, the PSNR gains of RCAN over EDSR are 0.49 dB and 0.55 dB. EDSR has much larger number of parameters (43 M) than ours (16 M), but our RCAN obtains much better performance.”
ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (2023, methodology): “It comprises 35 LR images widely used in recent literature [19, 58, 59, 60, 61].”
Enhancing Sketch-Based Image Retrieval by CNN Semantic Re-ranking (2020, methodology): “Besides, Matsui et al. [52] devised an SBIR re-ranking method, which is also based on explicit feed-back, for the search of Japanese comics.”
Activating More Pixels in Image Super-Resolution Transformer (2022, methodology): “Five benchmark datasets including Set5 [2], Set14 [66], BSD100 [40], Urban100 [19] and Manga109 [41] are used to evaluate the methods.”
Residual Non-local Attention Networks for Image Restoration (2019, background): “As shown in Table 6, our RNAN+ achieves the second best performance among benchmark datasets: Set5 (Bevilacqua et al., 2012), Set14 (Zeyde et al., 2010), B100 (Martin et al., 2001), Urban100 (Huang et al., 2015), and Manga109 (Matsui et al., 2017).”
Dual Reconstruction Nets for Image Super-Resolution with Gradient Sensitive Loss (2018, methodology): “We compare the performance of the proposed method with several state-of-the-art methods on ﬁve benchmark datasets, including SET5 [6], SET14 [49], BSDS100 [2], URBAN100 [17] and MANGA109 [32].”
Toward Real-World Single Image Super-Resolution: A New Benchmark and a New Model (2019, methodology): “There are several popular datasets, including Set5 [4], Set14 [59], BSD300 [32], Urban100 [19], Manga109 [33] and DIV2K [45] that have been widely used for training and evaluating the SISR methods.”
Efficient and Explicit Modelling of Image Hierarchies for Image Restoration (2023, neutral): “Method Scale # Params [M] Set5 [3] Set14 [78] BSD100 [49] Urban100 [28] Manga109 [50] PSNR↑ SSIM↑ PSNR↑ SSIM↑ PSNR↑ SSIM↑ PSNR↑ SSIM↑ PSNR↑ SSIM↑”
Residual Feature Aggregation Network for Image Super-Resolution (2020, neutral): “For testing, we use ﬁve standard benchmark datasets: Set5 [1], Set14 [33], B100 [20], Urban100 [12], and Manga109 [21].”
SwinIR: Image Restoration Using Swin Transformer (2021, methodology): “For ablation study, we train SwinIR on DIV2K [1] for classical image SR (×2) and test it on Manga109 [60].”
Dual Aggregation Transformer for Image Super-Resolution (2023, methodology): “We use five datasets for evaluation: Set5 [1], Set14 [26], B100 [19], Urban100 [11], and Manga109 [20] with three factors: × 2, × 3, and × 4.”
Feedback Network for Image Super-Resolution (2019, methodology): “We evaluate SR results under PSNR and SSIM[39] metrics on five standard benchmark datasets: Set5[3], Set14[41], B100[24], Urban100[15], and Manga109[25].”
Closed-Loop Matters: Dual Regression Networks for Single Image Super-Resolution (2020, methodology): “We compare different methods on five benchmark datasets, including SET5 [3], SET14 [47], BSDS100 [1], URBAN100 [21] and MANGA109 [29].”