DOAFT: Infrared and Visual Image Registration Based on Dominant Orientation Align Feature Transform

Shentao Wang,Peili Xi,Yun-ze Cai
DOI: https://doi.org/10.23919/chicc.2019.8865405
2019-01-01
Abstract:The registration of multi-modal images is a challenging problem. The registration of infrared and visual images is also one of them and is often encountered. To solve this problem, this paper proposes a method called dominant orientation align feature transform(DOAFT). The first problem we solve is how to align the dominant orientation of the image. We propose a method based on deep leaming, which is called perceptual similarity judgment to keep the rotation invariance first. Then we use the speed up robust feature(SURF) to implement the feature point detection, the maximum index map(MIM) to extract features and the robust point matching(RPM) to remove anomalies. Experimental results show that the proposed dominant orientation alignment method is effective. The final experimental results of registration on infrared and visual images are shown to outperform the best existing methods on the dataset we use.
What problem does this paper attempt to address?