Evaluation of the influence of cloud contamination in satellite image matching by the Phase-Only-Correlation method

If the Phase-Only-Correlation (POC) method, known as an accurate and robust algorithm for image matching, is applied to a satellite image with clouds, only clear areas after cloud screening can be input. However, if the cloud screening is not perfect, the method will be applied to a cloud-contaminated image. For example, it is difficult to perform accurate cloud screening for images created using JAXA's ALOS-2/Compact Infrared Camera (CIRC) with only a single thermal infrared (TIR) band, and Moderate Resolution Imaging Spectroradiometer (MODIS) cloud mask products are less accurate for some types of surfaces. However, the robustness of the POC method in cloud-contaminated satellite images has not been fully investigated. Thus, in the present paper, we evaluated it, particularly for TIR images, using two approaches. In the first approach, we placed various simulated clouds on actual CIRC images, and evaluated the accuracy of the POC method under cloud conditions that varied in terms of coverage and position. As a result, the gap estimation error was within 0.03 pixels, although the image-to-image similarity (alpha value) decreased and the error increased with increasing cloud coverage. This approach was also applied to Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) near-infrared (NIR) images and showed a similar tendency and a gap estimation error of 0.005 pixels or less. In the second approach, we used cloud-contaminated CIRC images as target images and ASTER/TIR images as base images, and the gaps between the CIRC and ASTER images were evaluated by the POC method. As a result, the gap estimation error was somewhat larger than that in the first approach, but was within 0.7 pixels. In conclusion, the POC method is robust against cloud contamination on satellite images if an error of 1 pixel is acceptable. This is useful information for some applications such as automatic satellite image registration and automatic geolocation validation.