Extraction of Latent Structure from Imaging Data Using Markov Random Field Models

2016 Fiscal Year Final Research Report

• Project/Area Number: 25280090
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Soft computing
• Research Institution: The University of Tokyo
• Principal Investigator: Okada Masato 東京大学, 新領域創成科学研究科, 教授 (90233345)
• Co-Investigator(Kenkyū-buntansha): 永田 賢二 東京大学, 新領域創成科学研究科, 助教 (10556062) ; 桑谷 立 国立研究開発法人海洋研究開発機構, 地球内部物質循環研究分野, 研究員 (60646785) ; 赤井 一郎 熊本大学, パルスパワー科学研究所, 教授 (20212392)
• Project Period (FY): 2013-04-01 – 2017-03-31
• Keywords: 画像処理 / ベイズ推論 / イメージング / 潜在構造 / ビッグデータ

Outline of Final Research Achievements

The purpose of this research was to propose an algorithm to extract latent structure from imaging data. As an initial objective, we proposed an algorithm that automatically estimates filter width using hyper-parameter of MRF model, clarified correspondence between MRF model and reaction diffusion equation, and proposed algorithm to estimate hyper-parameter distribution. Furthermore, we developed the method to analytically evaluate distribution estimation, and evaluated the influence of downsampling of the image data.
We applied extraction of latent structure to analysis using MRF model of data in Earth and geological sciences. We also applied extraction of latent structure from the STM / STS data of the physical science, using LASSO which is a method based on sparse modeling.

Free Research Field

ソフトコンピューティング