Registrating oblique images by integrating affine and scale-invariant features

ABSTRACT

Automatic registration of oblique images can be challenging due to the complexly geometric deformations of the images. A feature-based registration method for oblique images is proposed in this study. The proposed method integrates affine and scale-invariant features and mainly includes three steps: initial matching, propagative matching, and final registration. In the initial matching, the maximally stable extremal region (MSER) keypoints are detected and matched based on Scale-Invariant Feature Transform (SIFT) descriptors. The SIFT keypoints in the supporting region (SR) of MSER are matched using affine invariant normalized cross-correlation matching algorithm. Neighbourhood supporting strength – the ratio of SIFT matches to SIFT keypoints in SR, is proposed to eliminate error matches. In propagative matching, the matches in the former step are used for coarse geometric transformation estimation. The keypoints that have not been successfully matched in the initial matching are handled by iterative correspondence identification under geometric constraint. The two matching processes obtain evenly distributed and sufficient number of correspondences to calculate the accurate transformation between inputted images. Final registration is achieved using bilinear interpolation on sensed image. Experimental results carried on close-range, satellite, and aerial images demonstrate that the proposed method can achieve reliable correspondences and performs better than state-of-the-art methods for oblique image registration.

Acknowledgements

This study is supported by the National Natural Science Foundation of China: [Grant Number 41371438], the Graduate Scientific Research Innovation Program of Jiangsu Province Ordinary University [Grant Number KYLX16_0543].

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China: [Grant Number 41371438] and Graduate Scientific Research Innovation Program of Jiangsu Province Ordinary University: [Grant Number KYLX16_0543].