Region-based image registration for remote sensing imagery

Area of Science:

• Geosciences
• Computer Science
• Remote Sensing

Background:

• Accurate registration of remote sensing imagery is crucial for various geospatial applications.
• Traditional local feature-based methods can be susceptible to errors from feature estimators.

Purpose of the Study:

• To develop an automatic region-based registration method for remote sensing images.
• To improve registration accuracy by matching region properties instead of local features.

Main Methods:

• Automated image segmentation using kernel density estimators, morphological reconstruction, and watershed transform.
• Calculation of regional Fourier descriptors and standardized regional intensity descriptors.
• Definition of a joint matching cost based on Euclidean distances for region correspondence.

Main Results:

• The proposed joint intensity-Fourier descriptor method achieved high accuracy on synthetic and real datasets.
• Average root-mean-squared error (RMSE) of 0.446 ± 0.359 pixels and 1.152 ± 0.488 pixels on satellite imagery.
• Demonstrated lower registration error compared to Harris, FAST, SURF, BRISK, and KAZE keypoint descriptors.

Conclusions:

• The proposed region-based registration method offers high accuracy for remote sensing imagery.
• The joint intensity-Fourier descriptor approach effectively addresses limitations of local feature estimators.
• This technique provides a robust and accurate solution for remote sensing image registration.