HDR video generation based on dynamic mode decomposition and its applications

2021 Fiscal Year Final Research Report

• Project/Area Number: 18K18073
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: The University of Kitakyushu (2020-2021); Kyushu Institute of Technology (2019); Kagawa University (2018)
• Principal Investigator: Matsuoka Ryo 北九州市立大学, 国際環境工学部, 准教授 (40780391)
• Project Period (FY): 2018-04-01 – 2022-03-31
• Keywords: 高ダイナミックレンジ画像 / スパースコーディング / 画像復元

Outline of Final Research Achievements

When taking a photograph in a high dynamic range (HDR) scene, the accuracy of object recognition is significantly degraded due to under/over-exposed pixels. This can be solved by using an HDR sensor that faithfully preserves the optical information of the scene, but it is very expensive. Therefore, image-blending-based HDR camera technology using a low dynamic range sensor is required. However, existing image blending techniques require huge computational costs and large computational resources, and many challenges remain for practical applications.
In this research project, we proposed an efficient and practical HDR video generation technique. In addition, we developed reconstruction algorithms based on sparse modeling to achieve high-quality HDR video restoration.

Free Research Field

画像処理

Academic Significance and Societal Importance of the Research Achievements

本研究の主要な成果は, 既存の画像統合型HDR動画像生成技術に比べて低計算量でかつ効率的なHDR動画像の復元が可能な手法であるため, 自動運転技術や監視カメラ技術の飛躍的な発展が期待できる. さらに, HDR動画像が重要な役割を果たす産業・サイエンス・工学の諸分野に大きく貢献するものである. また, 計算量の観点から実現が困難であった画期的な動画像処理応用を切り開き, 医療・農業・産業などの広範な分野に大きなインパクトを与えるものであると考えられる.