Precision Geometric Correction of Very-High-Resolution Satellite Imagery Using Multi-Resolution GCP Chip Matching Hyeona Kim (a), Taejung Kim (b*)
a) Program in Smart City Engineering, Inha University
100 Inha-ro, Incheon 22212, Republic of Korea
b) Dept. of Geoinformatic Engineering, Inha University
100 Inha-ro, Incheon 22212, Republic of Korea
*tezid[at]inha.ac.kr
Abstract
With advancements in satellite image processing technologies and the expansion of various remote sensing applications, the demand for very high-resolution satellite imagery is increasing. To effectively utilize satellite image products across diverse fields, it is essential to eliminate geometric distortions of raw imagery through precise geometric correction and ensure consistent positional accuracy. To this end, a Ground Control Point (GCP) chip database, which integrates accurate ground coordinates and image patches, is constructed and automatically matched with satellite images to establish a precision sensor model. The performance of the matching and the accuracy of sensor model is influenced by the resolution of both the satellite imagery and the GCP chips. This study performs geometric correction of very high-resolution satellite imagery using GCP chips of different resolutions and evaluates the performance of the resulting precision sensor model. GCP chips and satellite images are upsamped to meet their spatial resolution. The upsampled chips and images are matched against each other to generate GCPs automatically. Different upsampling ratios were applied for matching, with bicubic interpolation used during the upsampling process. The absolute positional accuracy of the final orthoimage was evaluated to confirm that it met the level required for practical applications. The experiment used WorldView-3 satellite imagery with a ground sampling distance of 30cm and GCP chips at multiple spatial resolutions. Accuracy was assessed based on model point error, checkpoint error, and mapping error. The results showed the lowest errors when high-resolution UAV chips were used. With upsampling, the checkpoint pixel error improved by approximately 62 percent, and the orthoimage mapping error decreased by about 40 percent compared to raw-resolution matching. The final checkpoint and mapping errors remained within 1.5 pixels and 0.5 meters, respectively, demonstrating that the proposed method was applicable for generating very high-precision image maps automatically.
