Presentation Information

1:30 PM - 1:45 PM JST(4:30 AM - 4:45 AM UTC)Candidate for Academic Incentive AwardCareer Explorer

[A-8-14]RGB Image Restoration Based on Image Features for Sensing Image Sensors

◎△Kosuke Nakamura¹, Keita Kitagawa¹, Kota Yoshida², Yohei Hori², Shunsuke Okura¹ (1. Ritsumeikan Univ., 2. AIST)

Keywords:

Sensing Image Sensor,Image Restoration,Feature Image,Horizontal Edge Feature,U-Net,Color Reproducibility

This paper proposes an RGB image restoration method based on horizontal edge features extracted from each RGB channel for sensing image sensors. In conventional one-channel feature-based methods, color information is lost due to grayscale conversion, which causes problems in the color reproducibility of restored images. To address this issue, the proposed method preserves the edge information of each RGB channel as three-channel features and restores RGB color images using U-Net. Evaluation using the INRIA Person Dataset confirmed that the proposed method improves SSIM, LPIPS, and CIEDE2000 compared with the conventional method, thereby enhancing both structural and color reproducibility.

Back to Session information