Abstract



The mid-wave infrared (MIR) sensor onboard KOMPSAT-3A captures thermal imagery within the 3.3–5.2 μm spectral range, enabling detailed thermal analysis under both daytime and nighttime conditions. These images are extensively utilized in various applications, including urban heat island monitoring, drought assessment, environmental analysis, and thermal anomaly detection. However, temporal discrepancies between daytime and nighttime acquisitions frequently introduce radiometric inconsistencies and relative geometric dissimilarities, which pose significant challenges for accurate image registration. To address these issues, this study proposes a two-stage image registration framework that integrates radiometric normalization and feature-based alignment. In the first stage, gamma correction and Box-Cox transformation are employed to mitigate radiometric discrepancies, thereby improving feature reliability. In the second stage, the robust invariant feature transform (RIFT) algorithm is enhanced by incorporating Gaussian weighting, which refines the phase congruency (PC) map, leading to more robust feature extraction under varying conditions. Feature correspondences are filtered using the random sample consensus (RANSAC) algorithm to remove outliers, ensuring reliable matching. An affine transformation model is estimated using inlier points to align nighttime MIR imagery with daytime reference. The proposed framework, integrating radiometric and feature contrast enhancement methods, was evaluated across four distinct geographic sites. Experimental results demonstrated significant improvements over conventional methods, reducing relative geometric dissimilarities and enhancing image registration accuracy.





Keywords 

KOMPSAT-3A, Mid-wave infrared imagery, Image registration