Research on rapid and accurate image restoration by fusion of signal processing and deep learning

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K04472
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 21020:Communication and network engineering-related
• Research Institution: Keio University
• Principal Investigator: Ikehara Masaaki 慶應義塾大学, 理工学部(矢上), 教授 (00212796)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 画像復元 / 深層学習 / 信号処理

Outline of Final Research Achievements

In this research project, we developed a high-speed, high-precision image restoration technique by combining signal processing and deep learning. By using signal processing to detect image features as a pre-processing step and deep learning while taking into account the characteristics of degraded images, we have achieved better performance than conventional methods for 1) fog removal, 2) raindrop removal, 3) video super-resolution, and 4) blur removal. Moreover, for convolution, which is a fundamental process in deep learning, we were able to construct a high-performance network while reducing the amount of operations by preparing several kernels and expressing the convolution by linear combination of the kernels.

Free Research Field

画像処理

Academic Significance and Societal Importance of the Research Achievements

本研究では、畳み込み層の役割を解明し、信号処理技術により目的に応じた画像の特徴を前処理として検出し、劣化画像を考慮しながら深層学習を行うことにより、より層数の少ないネット構造を用いて、高精度な画像復元技術を開発する。本研究によりカメラ撮影における劣化を低減させるのみならず、画像をセンサーとして用いるセキュリティー分野や社会のあらゆる領域を変革するブレークスルーになり得る。