Multispectral remote sensing image registration based on maximally stable extremal regions
Jun Feng Guo, Hao Sun, Changren Zhu, Shunping Xiao · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2009
Image registration is a vital step in the processing of multispectral remotesensing imagery. This paper presents a robust multispectral remotesensing image registration algorithm based on maximally stable extremal regions (MSERs). Firstly, MSERs are detected independently in the reference image and the sensed image. Secondly, the SIFT descriptor is adopted to capture texture information in the detected regions, while an affine invariant shape descriptor for MSER is constructed to ensure that features can be reliably matched regardless of the appearance change. Both the SIFT descriptors and the shape descriptors are matched using the Euclidean distance measurement. Matching results are then combined and the optimal corresponding points are chosen to estimate the transformation parameters. Finally, random sample consensus (RANSAC) algorithm is applied for geometry estimation. Experimental results on various image pairs demonstrate that the proposed MSER based algorithm is very effective for multispectral remotesensing image registration.