<p>Unmanned aerial vehicle (UAV) multispectral imagery enables centimeter-scale observation of coastal waters, yet sun glint frequently causes reflectance overestimation and introduces localized spikes and spatial discontinuities in orthomosaics. The problem is amplified in benthic-influenced shallow waters, where strong depth and benthic variability and spatially heterogeneous water constituents weaken the stable visible-near-infrared (VIS-NIR) coupling assumed by conventional regression-based glint-correction methods. We propose a Multi-View Glint Correction (MVGC) method that exploits the dense multi-view redundancy inherent to high-overlap UAV surveys. For each ground location, MVGC analyzes a multi-view observation stack and combines (i) NIR-distribution-based outlier rejection to retain low-glint candidates and (ii) spectral-angle screening using the Spectral Angle Mapper (SAM) to exclude view-dependent spectral distortions, before aggregating only the selected candidates to reconstruct corrected reflectance. Quantitative evaluation showed that MVGC substantially weakened glint-driven VIS-NIR coupling relative to Joyce-style baselines. Regression slopes were reduced by 86% in the blue band (0.42–0.06), 68% in the green band (0.34–0.11), and 75% in the red band (0.28–0.07), while the corresponding R² values decreased from 0.71, 0.60, and 0.67 to 0.06, 0.11, and 0.17, respectively, indicating markedly weaker residual band-coupled glint signal. Over homogeneous water patches, MVGC achieved the strongest suppression of extremes and dispersion among the compared methods, reducing the blue-band maximum to 0.0065 sr ⁻¹ and the blue-band standard deviation to 0.0002 while preserving plausible mean reflectance levels. Along-transect cross-sectional profiles corroborated these findings: spike-like anomalies exceeding 0.01 sr ⁻¹ evident in baseline products were largely eliminated under MVGC, yielding smoother and more spatially continuous band-wise reflectance fields. Overall, MVGC reframes glint correction as a location-specific selection-and-aggregation problem over multi-view stacks, providing a more stable reflectance product in shallow, heterogeneous coastal waters than single-view global regression-based correction applied to conventional orthomosaics.</p>