多光谱图像
双线性插值
亚像素渲染
脱模
插值(计算机图形学)
计算机科学
欠采样
人工智能
计算机视觉
图像分辨率
采样(信号处理)
混叠
像素
全光谱成像
多元插值
空间分析
双三次插值
滤波器(信号处理)
遥感
多光谱模式识别
高光谱成像
算法
图像缩放
图像传感器
空间滤波器
光谱带
最近邻插值
空间相关性
能见度
图像处理
作者
Jianfeng Yao,Kacoutchy Jean Ayikpa,Pierre Gouton,Koné Tiémoman
标识
DOI:10.1016/j.optcom.2025.132842
摘要
Multispectral imaging enables detailed material analysis by capturing multiple spectral bands. Conventional systems acquire each band at full resolution but are costly and slow. Compact one-shot multispectral filter array (MSFA) sensors offer a practical alternative, capturing subsampled spectral mosaics that require demosaicking to recover full-resolution images. However, traditional spatial interpolation methods such as Weighted Bilinear (WB) and BTES often introduce blurring or directional artifacts in regions with strong spatial variation. To address these limitations, we introduce Triangular Structure-aware Bilinear Interpolation (TriSBI), a geometry-driven spatial method designed to better preserve edges and textured structures in MSFA mosaics. While TriSBI improves over WB and BTES, the severe undersampling inherent to one-shot MSFA sensors still restricts any spatial method’s ability to reconstruct sharp transitions and fine details. Motivated by this constraint, we adopt a multi-shot acquisition strategy with controlled subpixel shifts, increasing the sampling density of each spectral band. This enhanced sampling strengthens spatial correlations and allows TriSBI to deliver sharper reconstructions in contour-rich and highly textured regions. Beyond its standalone performance, TriSBI also provides a stronger spatial foundation for spectral–spatial demosaicking. Its integration into Spectral Difference (SD) and Iterative Spectral Difference (ISD) improves inter-band correlation modeling and yields clearer reconstructions of complex structures compared to WB-based baselines. Experiments on the CAVE and TokyoTech datasets show consistent improvements in PSNR, SSIM, and SAM for both one-shot and multi-shot configurations, demonstrating the effectiveness of combining geometry-aware spatial interpolation with enhanced acquisition strategies for MSFA-based multispectral imaging. • Introduction of a geometry-aware spatial interpolation method (TriSBI) for improved MSFA demosaicking. • TriSBI reduces blurring and artifacts compared to conventional Weighted Bilinear and BTES methods. • A multi-shot acquisition strategy is employed to mitigate spectral undersampling, enhancing spatial reconstruction. • Integrating TriSBI into SD/ISD spectral-spatial methods, particularly with multi-shot, significantly boosts multispectral image quality. • Experiments on realistic sensor-based projections of CAVE and TokyoTech datasets confirm gains in PSNR, SSIM, and SAM.
科研通智能强力驱动
Strongly Powered by AbleSci AI