Developing algorithms for fitting digitized geometric shapes to noisy data including outliers

2015 Fiscal Year Final Research Report

• Project/Area Number: 26540091
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Perceptual information processing
• Research Institution: National Institute of Informatics
• Principal Investigator: Sugimoto Akihiro 国立情報学研究所, コンテンツ科学研究系, 教授 (30314256)
• Research Collaborator: Kenmochi Yukiko University Pari-Est, CNRS researcher
• Project Period (FY): 2014-04-01 – 2016-03-31
• Keywords: 幾何形状 / あてはめ / 外れ値 / コンピュータビジョン

Outline of Final Research Achievements

Existing methods for fitting geometric shapes have a fatal drawback that data to be fitted are digitized while the shape model is continuous. In order to solve this problem, this research takes an approach where we fit a digitized geometric shape to given noisy data including outliers. We have developed methods for finding the 2D discrete polynomial curve that explains best given noisy data, for finding the 3D discrete polynomial curve that explains best given noisy data, and for finding the discrete rigid transformation that explains best for a given pair of digital images. All of our proposed methods were developed based on the same idea that formulated discrete optimization problems are tackled via the local search to find the (locally) optimal solution.

Free Research Field

情報学