Distance Set as a Local Shape Descriptor and its Improvement

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K00308
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: Hiroshima City University
• Principal Investigator: Iwata Kazunori 広島市立大学, 情報科学研究科, 准教授 (20405492)
• Project Period (FY): 2016-10-21 – 2019-03-31
• Keywords: 形状認識 / 形状検索 / 形状クラスタリング / 距離集合 / ヒルベルト距離 / フルプロクルステス距離 / 距離計量学習 / 通常プロクルステス二乗和

Outline of Final Research Achievements

A shape is formed of contours, such as line drawings by pen or boundaries obtained through image segmentation. The contours can be represented as a finite set of points using contour sampling. Such a finite set of points is called a sample set.
Considering the geometrical configuration of a sample set we presented several methods for shape classification. As an example of considering the geometrical configuration, the classification results do not depend on the similarity transformations to each shape, which consist of translation, rotation, and isotropic scaling. In this series of studies, matching shapes indicates determining the correspondence between their respective sample sets, and is consistently an underlying fundamental process in shape classification. Such matching-based approach gives several merits: for example, we can visually check the matching and clearly confirm the computational reasons for the classification results.

Free Research Field

パターン認識

Academic Significance and Societal Importance of the Research Achievements

研究成果で得られた方法は，手書き文字など特定の種類の形状に対して特化したようなものではなく，線描画や輪郭線なども統一して扱うことのできる用途の広いものである．また，形状の検索，分類，クラスタリングといった様々な解析の手段を与えている．
近年はスマートフォンやタブレットなどタッチパネルを備えた器機が急速に普及してきていることから，研究成果の応用例として，タッチパネルに指を使って描いた線描画をクエリーとしたインターネット上の画像検索システムの開発を考えている．