東華大學圖書館 |

Deep Understanding of Sketches and Comics.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Deep Understanding of Sketches and Comics./
作者:	Li, Chengze.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2020,
面頁冊數:	185 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-03, Section: A.
Contained By:	Dissertations Abstracts International83-03A.
標題:	Computer engineering. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28735910
ISBN:	9798535514727

Deep Understanding of Sketches and Comics.
Li, Chengze.

Deep Understanding of Sketches and Comics. - Ann Arbor : ProQuest Dissertations & Theses, 2020 - 185 p.

Source: Dissertations Abstracts International, Volume: 83-03, Section: A.

Thesis (Ph.D.)--Hong Kong University of Science and Technology (Hong Kong), 2020.

This item is not available from ProQuest Dissertations & Theses.

Recently, electronic devices have become a key component of our daily life. People are starting to enjoy cartoons and comics on e-devices with enriched reading experiences. These media on e-devices can be retargeted to arbitrary resolution to fit the user's screen resolution. The contents of these media can also be enriched with further editing of visual and sound effects. Despite the strong market needs, the production of comics and cartoons is still time-consuming and labor-intensive, as computers are unable to achieve the semantic understanding of these media and cannot facilitate the production pipeline. In this thesis, we bring the idea of deep visual analysis from the field of computer vision and graphics to perform deep semantic analysis of sketch and comics. We further propose several frameworks and applications that help benefit the comic and cartoon workflow.The first problem we address relates to the automatic structural line extraction from manga images, which is critical to migrate legacy manga to the digital domain. Unfortunately, it is very challenging to distinguish structural lines from arbitrary, highly-structured, and black-and-white screen patterns. We propose a novel data-driven approach to identify structural lines from the pattern-rich manga with no assumption of the screen patterns. The method is based on a tailored CNN model. We also develop an efficient and effective way to generate a rich set of training data pairs. We evaluate our method on a large number of mangas of various drawing styles, and results show our method suppresses arbitrary screen patterns, regardless of their scales and appearances.In our second work, we focus on the problem of sketch colorization. Unlike photo colorization that strongly relies on texture information, sketch colorization is more challenging due to its abstract nature. We propose to tackle this problem with a two-stage learning-based framework. In the first drafting stage, our model guesses color regions and splashes a rich variety of colors over the sketch to obtain a color draft. The second refinement stage detects the unnatural colors and artifacts and tries to fix and refine the result. Comparing to existing approaches, this two-stage design effectively divides the complex colorization task into two simpler and goal-clearer subtasks. This eases the learning and achieves high-quality artifact-free colorization. Our last work aims at the task of the automatic filling style conversion of comics. We identify that the major obstacle in the conversion stems from the difference between the fundamental properties of screened region-filling and colored region-filling. To resolve this obstacle, we propose a screentone variational autoencoder, ScreenVAE, to encode the screened manga to an intermediate domain, which summarizes local texture characteristics. The learning-based model effectively unifies the properties of screening and color-filling, and ease the learning for bidirectional translation between screened manga and color comics. We further propose a network to learn the translation between the intermediate domain and color comics. Our model can generate superior screened manga given a color comic, and generate color comic that retains the original screening intention by the bitonal manga artist.

ISBN: 9798535514727Subjects--Topical Terms:

621879
Computer engineering.
Subjects--Index Terms:

Deep understanding

Deep Understanding of Sketches and Comics.
LDR:04428nmm a2200361 4500 001 2348808
005 20220908125424.5
008 241004s2020 ||||||||||||||||| ||eng d
020 $a 9798535514727
035 $a (MiAaPQ)AAI28735910
035 $a AAI28735910
040 $a MiAaPQ $c MiAaPQ
100 1 $a Li, Chengze. $3 3688180
245 1 0 $a Deep Understanding of Sketches and Comics.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 185 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-03, Section: A.
500 $a Advisor: Fu, Chi Wing.
502 $a Thesis (Ph.D.)--Hong Kong University of Science and Technology (Hong Kong), 2020.
506 $a This item is not available from ProQuest Dissertations & Theses.
520 $a Recently, electronic devices have become a key component of our daily life. People are starting to enjoy cartoons and comics on e-devices with enriched reading experiences. These media on e-devices can be retargeted to arbitrary resolution to fit the user's screen resolution. The contents of these media can also be enriched with further editing of visual and sound effects. Despite the strong market needs, the production of comics and cartoons is still time-consuming and labor-intensive, as computers are unable to achieve the semantic understanding of these media and cannot facilitate the production pipeline. In this thesis, we bring the idea of deep visual analysis from the field of computer vision and graphics to perform deep semantic analysis of sketch and comics. We further propose several frameworks and applications that help benefit the comic and cartoon workflow.The first problem we address relates to the automatic structural line extraction from manga images, which is critical to migrate legacy manga to the digital domain. Unfortunately, it is very challenging to distinguish structural lines from arbitrary, highly-structured, and black-and-white screen patterns. We propose a novel data-driven approach to identify structural lines from the pattern-rich manga with no assumption of the screen patterns. The method is based on a tailored CNN model. We also develop an efficient and effective way to generate a rich set of training data pairs. We evaluate our method on a large number of mangas of various drawing styles, and results show our method suppresses arbitrary screen patterns, regardless of their scales and appearances.In our second work, we focus on the problem of sketch colorization. Unlike photo colorization that strongly relies on texture information, sketch colorization is more challenging due to its abstract nature. We propose to tackle this problem with a two-stage learning-based framework. In the first drafting stage, our model guesses color regions and splashes a rich variety of colors over the sketch to obtain a color draft. The second refinement stage detects the unnatural colors and artifacts and tries to fix and refine the result. Comparing to existing approaches, this two-stage design effectively divides the complex colorization task into two simpler and goal-clearer subtasks. This eases the learning and achieves high-quality artifact-free colorization. Our last work aims at the task of the automatic filling style conversion of comics. We identify that the major obstacle in the conversion stems from the difference between the fundamental properties of screened region-filling and colored region-filling. To resolve this obstacle, we propose a screentone variational autoencoder, ScreenVAE, to encode the screened manga to an intermediate domain, which summarizes local texture characteristics. The learning-based model effectively unifies the properties of screening and color-filling, and ease the learning for bidirectional translation between screened manga and color comics. We further propose a network to learn the translation between the intermediate domain and color comics. Our model can generate superior screened manga given a color comic, and generate color comic that retains the original screening intention by the bitonal manga artist.
590 $a School code: 1223.
650 4 $a Computer engineering. $3 621879
650 4 $a Fine arts. $3 2122690
650 4 $a Computer science. $3 523869
650 4 $a Software. $2 gtt. $3 619355
650 4 $a Artists. $3 524768
650 4 $a Accuracy. $3 3559958
650 4 $a Computers. $3 544777
650 4 $a Photographs. $3 627415
650 4 $a Manga. $2 swd $3 2048699
650 4 $a Regions. $3 3560048
650 4 $a Art. $3 516292
650 4 $a Methods. $3 3560391
650 4 $a Illustrations. $3 3683691
650 4 $a Digitization. $3 3681264
650 4 $a Algorithms. $3 536374
650 4 $a Ablation. $3 3562462
650 4 $a Visualization. $3 586179
650 4 $a Colorization. $3 3688181
650 4 $a Semantics. $3 520060
650 4 $a Editing. $3 601456
653 $a Deep understanding
653 $a Sketch
653 $a Comics
653 $a Manga
690 $a 0357
690 $a 0984
690 $a 0464
710 2 $a Hong Kong University of Science and Technology (Hong Kong). $3 1022235
773 0 $t Dissertations Abstracts International $g 83-03A.
790 $a 1223
791 $a Ph.D.
792 $a 2020
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28735910