Automatic Registration of Multi-source Remote Sensing Images Based on Region Growing

Ding NI,Hong-Bing MA

DOI: https://doi.org/10.3724/SP.J.1004.2014.01058

2014-01-01

Abstract:Multi-source remote sensing images are usually captured by different sensors, in different spectra and/or at different times, which makes them diffcult to match. Although a variety of methods have been proposed to solve this problem, most of them are only suitable for particular applications. It is still an open and challenging task to develop more stable and applicable algorithms. This paper presents a novel registration method based on region growing. It firstly utilizes the global matching to find seed points based on the updated scale-invariant features (SIFT), and then uses the seed points to start the region growing process. In the region growing phase, it estimates the transformation between the sensed image and the reference image employing the matching points in the current region, then expands the searching scope to find other matching points at farther places. This process iteratively executes until the searching region covers the whole image. Combined with bilateral matching, the proposed method can find a large number of evenly distributed matching points from only a few initial correct ones. Experiments show that this algorithm can find a greater number of matching points with higher precision than the existing methods for multi-source remote sensing images with significant gray-scale differences. Therefore, the proposed algorithm is more robust and powerful than several state-of-the-art methods.