CNN による漫画作者の画像的特徴の抽出

URI http://harp.lib.hiroshima-u.ac.jp/it-hiroshima/metadata/12336
File
Title
CNN による漫画作者の画像的特徴の抽出
Title Alternative
Image feature extraction of comic artist by CNN
Author
氏名 上田 智之
ヨミ ウエダ トモユキ
別名 UEDA Tomoyuki
氏名 小池 正記
ヨミ コイケ マサキ
別名 KOIKE Masaki
Subject
convolutional neural network
deep learning
image recognition
comic computing
Abstract

This paper reports the extraction of image features of comic artist by convolutional neural
network. In recent years, illegal uploading of comics has become a social issue. The comic artist
wants to find his own illegally uploaded comic. A convolutional neural network extract image
features representing a class from an image. Comics has the artist's style. The artist's style can
be used as an image features that represents a class in the comic image classification problem.
Therefore, convolutional neural networks can solve the artist classification problem for comic
images. If the classification problem can be solved, it can be applied to image retrieval. Visualize the
area of interest when convolutional neural networks predict artists. There is an element that can
predict the artist in the attention area. The experimental results suggest that the comic characters
face is important for convolutional neural network to predict the artist from the comic images.

Journal Title
広島工業大学紀要. 研究編
Volume
54
Spage
91
Epage
99
Published Date
2020-02
Publisher
広島工業大学
ISSN
1346-9975
NCID
AA11599110
Language
jpn
NIIType
Departmental Bulletin Paper
Text Version
出版社版
Rights
publisher
Set
it-hiroshima