Sketch-based manga retrieval using manga109 dataset

Yusuke MatsuiKota ItoYuji AramakiAzuma FujimotoToru OgawaToshihiko YamasakiKiyoharu Aizawa

article2015Multimedia tools and applications1,508 citations

Introduces Manga109 alongside a real-time, sketch-based retrieval framework that enables fast sub-image comic search through screentone-invariant edge features and interactive reranking.

Listen

As digital comic platforms grow, users face significant difficulty discovering content because commercial archives rely almost exclusively on basic text searches by title or author. This keyword-only approach fails to search visual content, which is problematic for Japanese comics (manga) that are defined by distinctive line art, non-textured graphics, and complex multi-frame page layouts. The article addresses this limitation by developing and evaluating an efficient, content-based retrieval system that enables users to search manga collections using hand-drawn sketches.

The system workflow preprocesses manga pages by identifying and filtering out blank inter-frame margins, then applies an object detector to locate candidate visual regions. To improve line detection, screentone shading patterns are removed before extracting edge orientation histogram features, which capture the geometric outlines of drawings regardless of scale. These visual features are compressed into compact binary codes using product quantization, enabling fast approximate nearest-neighbor matching against sketch queries. To support realistic testing, the researchers constructed Manga109, a publicly available benchmark dataset comprising 109 titles and 21,142 pages created by 94 professional artists.

Evaluation results show substantial improvements in both retrieval accuracy and processing efficiency. Across comparative tests, the proposed feature representation consistently outperformed standard image retrieval methods, achieving higher recall rates across varying visual targets. Feature compression enabled rapid querying, allowing the system to search across 14 million candidate regions from 21,142 pages in 70 milliseconds using 204 megabytes of memory on a single parallelized computer. Furthermore, interactive reranking methods, including relevance feedback (using retrieved images as new queries) and query retouching (editing existing sketches), allowed users to quickly refine searches to locate specific characters or visual traits across diverse titles.

These findings demonstrate that sketch-based retrieval is commercially viable at scale without requiring expensive computational infrastructure. Because the pipeline relies on compact indexing and lightweight vector matching, digital storefronts and archives can implement visual search on standard server hardware with minimal operational overhead. This offers an intuitive exploration mechanism for consumers while providing a standardized, legally cleared dataset to accelerate academic research in comic processing.

To build upon this work, developers should integrate text-filtering techniques to prevent Japanese text and speech balloons from appearing as false-positive sketch matches. Combining sketch queries with standard keyword metadata would also provide a hybrid search experience. While initial localization results indicate that finding small objects across millions of candidates remains challenging and subject to user drawing variability, the underlying architecture provides high confidence for large-scale, low-latency visual retrieval.

University of Tokyo color histogram high resolution image japanese sketch-based manga retrieval Manga109 dataset manga retrieval system

arXiv: 1510.04389

Paper: Aggregating Local Image Descriptors into Compact Codes, Hervé Jégou et al. (2012). It provides essential foundations for large-scale image indexing, descriptor aggregation, and product quantization used for scalable approximate nearest-neighbor search.
Paper: Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene Categories, Svetlana Lazebnik et al. (2006). It establishes spatial pyramid matching and edge orientation histogram representations that underlie spatial feature descriptions in visual retrieval.
Paper: Iterative Quantization: A Procrustean Approach to Learning Binary Codes for Large-Scale Image Retrieval, Yunchao Gong et al. (2013). It explores compact binary code learning and quantization strategies for efficient similarity search across large image collections.
Paper: Hamming Embedding and Weak Geometric Consistency for Large Scale Image Search, Hervé Jégou et al. (2008). It details indexing and geometric verification techniques in inverted file systems that inform fast visual search pipelines.
Paper: Content-Based Image Retrieval at the End of the Early Years, Arnold W.M. Smeulders et al. (2000). It surveys foundational concepts, feature extraction principles, and interactive relevance feedback in content-based image retrieval.
Paper: Shape Matching and Object Recognition Using Shape Contexts, Serge Belongie et al. (2002). It presents foundational shape-context and contour-matching methodologies critical for sketch-to-image alignment and retrieval.
Paper: Scalable Nearest Neighbor Algorithms for High Dimensional Data, Marius Muja et al. (2014). It introduces scalable approximate nearest-neighbor matching algorithms that underpin high-dimensional feature retrieval systems.
Paper: The use of MMR, diversity-based reranking for reordering documents and producing summaries, Jaime Carbonell et al. (1998). It defines classic interactive reranking and diversity principles that motivate sketch-based query refinement and feedback mechanisms.

Paper: Fine-Tuning CNN Image Retrieval with No Human Annotation, Filip Radenovic et al. (2017). It extends visual retrieval pipelines by training convolutional neural networks with generalized pooling and metric learning to replace hand-crafted edge descriptors.
Paper: Accelerating Large-Scale Inference with Anisotropic Vector Quantization, Ruiqi Guo et al. (2020). It advances vector quantization by introducing anisotropic loss for faster and more accurate nearest-neighbor search over high-dimensional image embeddings.
Paper: DiskANN: Fast Accurate Billion-point Nearest Neighbor Search on a Single Node, Suhas Jayaram Subramanya et al. (2019). It scales approximate nearest-neighbor indexing beyond in-memory product quantization to billion-point datasets on single-node SSD architectures.
Paper: Image-to-Image Translation with Conditional Adversarial Networks, Phillip Isola et al. (2017). It demonstrates conditional adversarial translation frameworks that advance cross-domain sketch-to-image synthesis and matching.

Cover for Sketch-based manga retrieval using manga109 dataset

Abstract

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

Topics

color histogram high resolution image japanese keyword-based search nearest neighbor search pseudo relevance feedback quantization image retrieval systems retrieval performance

Keywords

sketch-based manga retrieval Manga109 dataset manga retrieval system sketch-based image retrieval nearest-neighbor search line drawings content-based image retrieval

I Introduction
II Related Work
II-A Manga image description
II-B Retrieval and localization
II-C Querying
II-D Manga
III Manga Retrieval System
III-A EOH feature description with object windows
III-B Feature compression by product quantization
III-C Search
III-D Skipping margins
IV Query interaction
V Manga dataset
VI Experimental Results
VI-A Comparative study
VI-B Localization evaluation
VI-C Large-scale qualitative study
VII Conclusion
References

Knowls

Knowl 1 — The Manga109 Dataset
definition
Manga109 is a publicly available benchmark dataset of professional Japanese black-and-white comic books designed for multimedia and image-processing research. It comprises 109 manga titles spanning 21,142 pages drawn by 94 professional creators, with original publication dates ranging from the 1970s to the 2010s. The collection covers a wide variety of genres, including humor, battle, romantic comedy, animal, science fiction, sports, historical drama, fantasy, romance, suspense, horror, and four-frame cartoons. The average resolution of the page images is $833 \times 1179$ pixels, accommodating multi-frame layouts. Each manga title contains an average of 194 pages.
Knowl 2 — Margin Labeling and Skipping for Manga Pages
algorithm
To avoid extracting uninformative features from non-content regions and interframe spaces (margins) in manga pages, margin areas are detected and skipped before candidate feature extraction.
Input: Manga page image $I$ , margin area threshold $\theta = 0.1$ , candidate patch $x$
Output: Boolean decision whether to retain patch $x$
Apply morphological erosion to the white areas of $I$ to thicken line strokes and close small interframe gaps
Apply connected-component labeling to all white-connected regions in the eroded image
Identify the margin label $L_{\text{margin}}$ as the most frequent connected-component label appearing along the outermost peripheral borders of the page
For candidate patch $x$ with total pixel area $S(x)$ :
    Compute the margin pixel area $U(x)$ within $x$ that carries label $L_{\text{margin}}$
    Compute the margin ratio $r = U(x) / S(x)$
    if $r < \theta$ then
        Keep candidate patch $x$ for feature description
    else
        Discard candidate patch $x$
Knowl 3 — Screentone-Removed Edge Orientation Histogram (EOH) Patch Representation
model/method
To extract robust line features invariant to scale and shading artifacts, manga pages undergo screentone removal followed by candidate region proposal and Edge Orientation Histogram (EOH) extraction:
1. Screentone Removal: Halftone screentone patterns are separated and eliminated from the page image in advance to isolate clean contour line drawings and eliminate high-frequency dot texture noise.
2. Region of Interest Proposal: Candidate bounding boxes are generated via Selective Search (which achieves an object detection Area Under the Curve of 0.814 at an Intersection-over-Union threshold of 0.5, outperforming sliding window at 0.762 and BING at 0.657). Vertically or horizontally elongated boxes are subdivided into overlapping square boxes, and patches with side lengths smaller than 100 pixels are discarded.
3. EOH Feature Extraction: Each retained square patch is partitioned into a spatial grid of $c \times c$ cells (with $c=8$ ). Edges in each cell are quantized into 4 orientation bins and normalized. The concatenated $4c^2$ -dimensional descriptor ( $4 \times 8^2 = 256$ dimensions) for the entire patch is renormalized. Integral images are used to compute cell histograms across arbitrary window scales in constant time. All-zero vectors are eliminated.
Knowl 4 — Approximate Nearest-Neighbor Manga Retrieval using Product Quantization
model/method
A manga page is represented as a collection of $N$ localized 256-dimensional EOH patch descriptors $\{x_n\}_{n=1}^N$ . To enable large-scale search within compact memory and low latency, each feature vector $x_n$ is compressed into a compact binary code $q(x_n)$ via Product Quantization (PQ).

Under PQ with $M=16$ subvectors, the 256-dimensional vector is divided into 16 subvectors of dimension 16, each quantized to one of 256 subcentroids using an 8-bit unsigned integer (uint8_t), resulting in an $8M$ -bit (16-byte) code per patch. The quantized page representation is: $\mathcal{X}_{PQ} = \left\{ q(x_n) \;\middle|\; \frac{U(x_n)}{S(x_n)} < 0.1, \; n \in \{1, \dots, N\} \right\}$ where $S(x_n)$ is the area of patch $x_n$ and $U(x_n)$ is its margin area.

Given an uncompressed query sketch EOH feature vector $y \in \mathbb{R}^{256}$ and a collection of $P$ manga pages where the $p$ -th page contains $N^p$ quantized features $\mathcal{X}_{PQ}^p = \{q(x_n^p)\}_{n=1}^{N^p}$ , the nearest matching patch $(p^*, n^*)$ is retrieved by minimizing Asymmetric Distance Computation (ADC): $(p^*, n^*) = \arg\min_{p \in \{1, \dots, P\}, \, n \in \{1, \dots, N^p\}} d_{AD}\left(y, q(x_n^p)\right)$ where $d_{AD}(y, q(x_n^p))$ calculates the approximate Euclidean distance between query vector $y$ and compressed code $q(x_n^p)$ using precomputed subvector lookup tables.
Knowl 5 — Interactive Reranking Schemes for Sketch-Based Manga Retrieval
model/method
To refine search results beyond an initial freehand drawing, two interactive reranking mechanisms are employed:
1. Relevance Feedback: A user selects any target region or candidate object thumbnail from a retrieved manga page result. The system extracts the EOH descriptor from that professional artwork patch and re-executes the retrieval query in real time, allowing users to search without drawing complex shapes manually.
2. Query Retouch: A user directly modifies an active query (either their drawn sketch or an artist patch imported via relevance feedback) by adding strokes (e.g., drawing accessories like glasses) or erasing segments with an eraser tool. The retrieval results update automatically upon the completion of each stroke.
The feature extraction and matching pipeline is designed to be scale-invariant (by mapping patches of arbitrary sizes to a uniform $8 \times 8$ grid), rotation-variant (to suppress upside-down matches), and flip-variant (to preserve drawn directionality while allowing manual horizontal flipping if desired).

Knowl 6 — Comparative Retrieval Performance on Cropped Manga Frames

data/table

Retrieval performance was evaluated on a benchmark of 8,889 cropped frames taken from 10 manga titles in Manga109 across three target classes representing different difficulties: "Boy-with-glasses" (easy, 67 ground-truth instances), "Chombo" (normal, 178 ground-truth instances), and "Tatoo" (hard, 81 ground-truth instances). Ten participants (7 novices, 3 skilled artists) generated 30 query sketches in total (10 per target).

Method	Boy-with-glasses (Recall@100)	Chombo (Recall@100)	Tatoo (Recall@100)
Random Guess	0.0116	0.0114	0.0116
Compact OCM	0.0179	0.0073	0.0284
Bag-of-Features (BoF + SIFT)	0.0224	0.0185	0.0025
Fisher Vector (FV)	0.0313	0.0169	0.0000
Proposed (EOH + Screentone Removal)	0.0821	0.0174	0.0407

The proposed method achieved the highest Recall@100 across all targets. Bag-of-Features and Fisher Vector methods scored near zero on small targets such as "Tatoo" because global frame descriptors fail to capture localized sub-frame instances embedded among background clutter.

Knowl 7 — Impact of Spatial Grid Resolution and Product Quantization on Retrieval Recall
empirical result
Evaluating variations in spatial cell resolution ( $c \times c$ for $c \in \{4, 8, 16, 32\}$ ) and Product Quantization compression parameters reveals:
1. Spatial Grid Divisions: Retrieval recall does not increase monotonically with finer cell grids. Mean Recall@100 peaks at $c = 8$ ( $8 \times 8 = 64$ cells, 256 dimensions) and drops for finer divisions ( $16 \times 16$ and $32 \times 32$ ) as well as coarser divisions ( $4 \times 4$ ). Finer grids over-constrain the match, whereas moderate abstraction is required to bridge the geometric differences between rough user sketches and finished drawings.
2. PQ Compression Level: Compressing the 256-D EOH descriptor into $M$ subvectors reduces recall relative to uncompressed descriptors (mean Recall@100 drops from $\approx 0.082$ uncompressed to $\approx 0.057$ at $M=16$ , and $\approx 0.049$ at $M=8$ ). Setting $M=16$ provides an effective operational trade-off between memory footprint (16 bytes per candidate patch) and retrieval accuracy.

Knowl 8 — Manga Object Localization Performance and Scalability Benchmark

data/table

Object localization performance was evaluated using 30 "Boy-with-glasses" query sketches on two datasets: Lovehina (192 pages, 69 ground-truth annotated windows) and the complete Manga109 dataset (21,142 pages). Bounding boxes $B_p$ from Selective Search proposals filtered by margin skipping ( $U/S < 0.1$ ) were judged correct if the Intersection-over-Union overlap with ground-truth box $B_{gt}$ satisfied $\frac{\text{area}(B_p \cap B_{gt})}{\text{area}(B_p \cup B_{gt})} > 0.5$ .

Dataset	Pages	Patches	mAP@100	Memory	Runtime (Single-thread)
Lovehina	192	138,000	$1.12 \times 10^{-2}$	2.21 MB	11.6 ms
Manga109	21,142	14,000,000	$1.43 \times 10^{-4}$	204 MB	331 ms

When using a multi-threaded parallel implementation where candidate distances are scanned across manga titles simultaneously, retrieval latency over the entire 21,142 pages (14 million candidate patches) drops from 331 ms to 70 ms on a single desktop PC, requiring 204 MB of RAM.

Knowl 9 — Susceptibility to False Positive Retrieval on Manga Text Regions
limitation
Because Japanese manga pages contain numerous speech balloons, sound effects (onomatopoeia), and vertical/horizontal text regions with high edge densities and varied stroke orientations, the EOH descriptor often matches text areas to query sketches. The absence of an explicit text detection and filtering stage causes text-dense patches to be ranked among the top search results, reducing retrieval precision.

Coverage note — GUI software implementation details (such as thumbnail layout arrangement and preview window controls) and specific visual gallery examples from qualitative user queries have been omitted, as they represent interface presentation rather than core methodological or empirical contributions.

References

1.Alexe B, Deselaers T, Ferrari V (2012) Measuring the objectness of image windows. IEEE TPAMI 34(11):2189–2202
2.Aramaki Y, Matsui Y, Yamasaki T, Aizawa K (2014) Interactive segmentation for manga. In: Proceedings SIGGRAPH Poster. ACM
3.Babenko A, Lempitsky V (2014) The inverted multi-index. IEEE TPAMI
4.Buttcher S, Clarke CLA, Cormack GV (2010) Information Retrieval: Implementing and Evaluating Search Engines The MIT Press
5.Cao Y, Chan AB, Lau RWH (2012) Automatic stylistic manga layout. In: Proceedings SIGGRAPH Asia. ACM
6.Cao Y, Lau RWH, Chan AB (2014) Look over here: Attention-directing composition of manga elements. In: Proceedings SIGGRAPH. ACM
7.Cao Y, Wang C, Li Z, Zhang L, Zhang L (2010) Spatial-bag-of-features. In: Proceedings CVPR. IEEE
8.Cao Y, Wang C, Zhang L, Zhang L (2011) Edgel index for large-scale sketch-based image search. In: Proceedings CVPR. IEEE
9.Chen T, Cheng MM, Tan P, Shamir A, Hu SM (2009) Sketch2photo: Internet image montage. In: Proceedings SIGGRAPH Asia. ACM
10.Cheng MM, Zhang Z, Lin WY, Torr P (2014) Bing: Binarized normed gradients for objectness estimation at 300fps. In: Proceedings CVPR. IEEE
11.Chu WT, Chao YC (2014) Line-based drawing style description for manga classification. In: Proceedings MM. ACM
12.Chum O, Philbin J, Sivic J, Isard M, Zisserman A (2007) Total recall: Automatic query expansion with a generative feature model for object retrieval. In: Proceedings ICCV. IEEE
13.Dalal N, Triggs B (2005) Histograms of oriented gradients for human detection. In: Proceedings CVPR. IEEE
14.Delalandre M, Iwata M, Kise K (2014) Fast and optimal binary template matching application to manga copyright protection. In: Proceedings DAS
15.Eitz M, Hays J, Alexa M (2012) How do humans sketch objects? In: Proceedings SIGGRAPH. ACM
16.Eitz M, Hildebrand K, Boubekeur T, Alexa M (2011) Sketch-based image retrieval: Benchmark and bag-of-features descriptors. IEEE TVCG 17(11):1624–1636
17.Everingham M, Gool LV, Williams CKI, Winn J, Zisserman A (2010) The pascal visual object classes (voc) challenge. IJCV 88(2):303–338
18.Ge T, He K, Ke Q, Sun J (2014) Optimized product quantization. IEEE TPAMI 36(4):744–755
19.Guerin C, Rigaud C, Mercier A, Ammar-Boudjelal F, Betet K, Bouju A, Burie JC, Louis G, Ogier JM, Revel A (2013) eBDtheque: A Representative Database of Comics. In: Proceedings ICDAR. IEEE
20.Hoashi K, Ono C, Ishii D, Watanabe H (2011) Automatic preview generation of comic episodes for digitized comic search. In: Proceedings MM. ACM
21.Hu R, Barnard M, Collomosse J (2010) Gradient field descriptor for sketch based retrieval and localization. In: Proceedings ICIP. IEEE
22.Hu R, Wang T, Collomosse J (2011) A bag-of-regions approach to sketch-based image retrieval. In: Proceedings ICIP. IEEE
23.Ito K, Matsui Y, Yamasaki T, Aizawa K (2015) Separation of line drawings and screentones of manga. In: Proceedings Eurographics
24.Jégou H, Douze M, Schmid C (2008) Hamming embedding and weak geometric consistency for large scale image search. In: Proceedings ECCV
25.Jégou H, Douze M, Schmid C (2009) Packing bag-of-features. In: Proceedings ICCV. IEEE
26.Jégou H, Douze M, Schmid C (2011) Product quantization for nearest neighbor search. IEEE TPAMI 33(1):117–128
27.Jégou H, Perronnin F, Douze M, Sánchez J, Pérez P, Schmid C (2012) Aggregating local image descriptors into compact codes. IEEE TPAMI 34(9):1704–1716
28.Kalantidis Y, Avrithis Y (2014) Locally optimized product quantization for approximate nearest neighbor search. In: Proceedings CVPR. IEEE
29.Kazi RH, Igarashi T, Zhao S, Davis RC (2012) Vignette: Interactive texture design and manipulation with freeform gestures for pen-and-ink illustration. In: Proceedings CHI. ACM
30.King DE (2009) Dlib-ml: A machine learning toolkit. J Mach Learn Res 10:1755–1758
31.Kopf J, Lischinski D (2012) Digital reconstruction of halftoned color comics. In: Proceedings SIGGRAPH Asia. ACM
32.Lab AY (2015) Manga109 dataset. http://www.manga109.org
33.Lazebnik S, Schmid C, Ponce J (2006) Beyond bags of features: Spatial pyramid matching for recognizing natural scene categories. In: Proceedings CVPR. IEEE
34.Le TN, Luqman MM, Burie JC, Ogier JM (2015) A comic retrieval system based on multilayer graph representation and graph mining. In: Graph-Based Representations in Pattern Recognition, LNCS, vol 9069, pp 355–364
35.Levi K, Weiss Y (2004) Learning object detection from a small number of examples: the importance of good features. In: Proceedings CVPR. IEEE
36.Lin G, Shen C, Shi Q, van den Hengel A, Suter D (2014) Fast supervised hashing with decision trees for high-dimensional data. In: Proceedings CVPR. IEEE
37.Lowe DG (2004) Distinctive image features from scale-invariant keypoints. IJCV 60(2):91–110
38.Marchand-Mailet S, Sharaiha YM (1999) Binary Digital Image Processing: A Discrete Approach Academic Press
39.Matsui Y., Aizawa K., Jing Y. (2014) Sketch2manga: Sketch-based manga retrieval. In: Proceedings ICIP. IEEE
40.Matsui Y, Yamasaki T, Aizawa K (2011) Interactive manga retargeting. In: Proceedings SIGGRAPH Poster. ACM
41.Mei T, Rui Y, Li S, Tian Q (2014) Multimedia search reranking: A literature survey. ACM Comp Surv 46(3):38:1–38:38
42.Ojala T, inen MP, Maenpaa T (2002) Multiresolution gray-scale and rotation invariant texture classification with local binary patterns. IEEE TPAMI 24(7):971–987
43.Olsen L, Samavati FF, Sousa MC, Jorge JA (2009) Sketch-based modeling: A survey. Comput Graph 33(1):85–103
44.Philbin J, Chum O, Isard M, Sivic J, Zisserman A (2007) Object retrieval with large vocabularies and fast spatial matching. In: Proceedings CVPR. IEEE
45.Philbin J, Chum O, Isard M, Sivic J, Zisserman A (2008) Lost in quantization: Improving particular object retrieval in large scale image databases. In: Proceedings CVPR. IEEE
46.Qu Y, Pang WM, Wong TT, Heng PA (2008) Richness-preserving manga screening. In: Proceedings SIGGRAPH Asia. ACM
47.Qu Y, Wong TT, Heng PA (2006) Manga colorization. In: Proceedings SIGGRAPH. ACM
48.Ren S, Jin C, Sun C, Zhang Y (2014) Sketch-based image retrieval via adaptive weightinge. In: Proceedings ICMR. ACM
49.Rigaud C, Burie JC, Ogier JM, Karatzas D, de Weijer JV (2013) An active contour model for speech balloon detection in comics. In: Proceedings ICDAR. IEEE
50.Rigaud C, Guérin C, Karatzas D, Burie JC, Ogier JM (2015) Knowledge-driven understanding of images in comic books. IJDAR 18(3):199–221
51.Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S, Huang Z, Karpathy A, Khosla A, Bernstein M, Berg AC, Fei-Fei L (2014) Imagenet large scale visual recognition challenge. CoRR abs/1409.0575
52.Sánchez J, Perronnin F, Mensink T, Verbeek JJ (2013) Image classification with the fisher vector: Theory and practice. IJCV 105(3):222–245
53.Sato K, Matsui Y, Yamasaki T, Aizawa K (2014) Reference-based manga colorization by graph correspondence using quadratic programming. In: Proceedings SIGGRAPH Asia Technical Brief. ACM
54.Schneider RG, Tuytelaars T (2014) Sketch classification and classification-driven analysis using fisher vectors. In: Proceedings SIGGRAPH Asia. ACM
55.Serra J (1983) Image Analysis and Mathematical Morphology Academic Press
56.Shotton J, Blake A, Cipolla R (2008) Multi-scale categorical object recognition using contour fragments. IEEE TPAMI 30(7):1270–1281
57.Sivic J, Zisserman A (2003) Video google: A text retrieval approach to object matching in videos. In: Proceedings ICCV. IEEE
58.Smeulders AW, Worring M, Santini S, Gupta A, Jain R (2000) Content-based image retrieval at the end of the early years. IEEE TPAMI 22(12):1349–1380
59.Spyromitros-Xioufis E, Papadopoulos S, Kompatsiaris IY, Tsoumakas G, Vlahavas I (2014) A comprehensive study over vlad and product quantization in large-scale image retrieval. IEEE TMM 16(6):1713–1728
60.Sun W, Burie JC, Ogier JM, Kise K (2013) Specific comic character detection using local feature matching. In: Proceedings ICDAR. IEEE
61.Sun W, Kise K (2013) Detection of exact and similar partial copies for copyright protection of manga. IJDAR 16(4):331–349
62.Sun X, Wang C, Xu C, Zhang L (2013) Indexing billions of images for sketch-based retrieval. In: Proceedings MM. ACM
63.Sýkora D, Dingliana J, Collins S (2009) Lazybrush: Flexible painting tool for hand-drawn cartoons. In: Proceedings Eurographics
64.Tanaka T, Shoji K, Toyama F, Miyamichi J (2007) Layout analysis of tree-structured scene frames in comic images. In: Proceedings IJCAI
65.Tseng KY, Lin YL, Chen YH, Hsu WH (2012) Sketch-based image retrieval on mobile devices using compact hash bits. In: Proceedings MM. ACM
66.Uijlings JRR, van de Sande KEA, Gevers T, Smeulders AWM (2013) Selective search for object recognition. IJCV 104(2):154–171
67.Vedaldi A, Fulkerson B (2010) Vlfeat: An open and portable library of computer vision algorithms. In: Proceedings MM
68.Wang J, Liu W, Kumar S, Chang SF (2016) Learning to hash for indexing big data - a survey. Proc IEEE 104(1):34–57
69.Wang J, Zhang T, Song J, Sebe N, Shen HT (2016) A survey on learning to hash. CoRR abs/1606.00185
70.Wang X, Bai X, Ma T, Liu W, Latecki LJ (2012) Fan shape model for object detection. In: proceedings CVPR. IEEE
71.Wu Z, Ke Q, Isard M, Sun J (2009) Bundling features for large scale partial-duplicate web image search. In: proceedings CVPR. IEEE
72.Yu Q, Yang Y, Song YZ, Xiang T, Hospedales TM (2015) Sketch-a-net that beats humans. In: proceedings BMVC
73.Zhang X, Liu W, Dundar M, Badve S, Zhang S (2015) Towards large-scale histopathological image analysis: Hashing-based image retrieval. IEEE TMI 34(2):496–506
74.Zhang Y, Jia Z, Chen T (2011) Image retrieval with geometry-preserving visual phrases. In: proceedings CVPR. IEEE
75.Zhou R, Chen L, Zhang L (2012) Sketch-based image retrieval on a large scake database. In: proceedings MM. ACM

Citation

MLA

Matsui, Y., et al. “Sketch-based Manga Retrieval Using Manga109 Dataset”. Multimedia Tools and Applications, vol. 76, no. 20, 2016, pp. 21811–38, https://doi.org/10.1007/s11042-016-4020-z.

APA

Matsui, Y., Ito, K., Aramaki, Y., Fujimoto, A., Ogawa, T., Yamasaki, T., & Aizawa, K. (2016). Sketch-based manga retrieval using manga109 dataset. Multimedia Tools and Applications, 76(20), 21811–21838. https://doi.org/10.1007/s11042-016-4020-z

Chicago

Matsui, Y., K. Ito, Y. Aramaki, et al. 2016. “Sketch-based Manga Retrieval Using Manga109 Dataset”. Multimedia Tools and Applications 76 (20): 21811–38. https://doi.org/10.1007/s11042-016-4020-z.

Harvard

Matsui, Y. et al. (2016) “Sketch-based manga retrieval using manga109 dataset”, Multimedia Tools and Applications, 76(20), pp. 21811–21838. Available at: https://doi.org/10.1007/s11042-016-4020-z.

Vancouver

1. Matsui Y, Ito K, Aramaki Y, Fujimoto A, Ogawa T, Yamasaki T, Aizawa K (2016) Sketch-based manga retrieval using manga109 dataset. Multimedia Tools and Applications 76:21811–21838

BibTeX

@article{Matsui_2016, title={Sketch-based manga retrieval using manga109 dataset}, volume={76}, ISSN={1573-7721}, url={http://dx.doi.org/10.1007/s11042-016-4020-z}, DOI={10.1007/s11042-016-4020-z}, number={20}, journal={Multimedia Tools and Applications}, publisher={Springer Science and Business Media LLC}, author={Matsui, Yusuke and Ito, Kota and Aramaki, Yuji and Fujimoto, Azuma and Ogawa, Toru and Yamasaki, Toshihiko and Aizawa, Kiyoharu}, year={2016}, month=Nov, pages={21811–21838} }

Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

Sketch-based manga retrieval using manga109 dataset

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — The Manga109 Dataset

Knowl 2 — Margin Labeling and Skipping for Manga Pages

Knowl 3 — Screentone-Removed Edge Orientation Histogram (EOH) Patch Representation

Knowl 4 — Approximate Nearest-Neighbor Manga Retrieval using Product Quantization

Knowl 5 — Interactive Reranking Schemes for Sketch-Based Manga Retrieval