Retrieval and edit of illustration images based on drawing styles

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K00151
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Multimedia database
• Research Institution: Toyohashi University of Technology
• Principal Investigator: Kuriyama Shigeru 豊橋技術科学大学, 工学(系)研究科(研究院), 教授 (20264939)
• Co-Investigator(Renkei-kenkyūsha): OHBUCHI Ryutarou 山梨大学, 大学院総合研究部, 教授 (80313782)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 画像検索 / イラスト画像 / 描画スタイル / スタイル識別子 / 特徴分類 / 教師無し機械学習 / 局所特徴量 / Fisher Vector

Outline of Final Research Achievements

This research proposed a block-wise local discriminator for drawing styles of illustration images. The bags-of-features for styles are coded to compose a dictionary, which is utilized to generate fisher vectors. We reduced their dimensions by introducing non-linear, non-supervised manifold learning for improving the performance.
Through the comparative evaluation of discrimination accuracy, we have found that our method has superior performance in discriminating styles, and we received six kinds of awards from both domestic academic society and international symposium, which shows the highly-evaluated contribution of our method.
Our style-discriminators can be utilized for the retrieval, visualization, and transfer of drawing styles to improve the design of illustrative drawings, for a large scale of image data set such as clip-art archives.

Free Research Field

計算基盤