Multiscale Template Matching for Multimodal Remote Sensing Image

Abstract

Multimodal matching remains a difficult and pressing problem in the imaging processing community. Accurate and robust multimodal matching is important for the performance of applications, such as registration and fusion. Traditional image matching algorithms cannot effectively handle multimodal images with severe nonlinear radiometric distortion (NRD). In this article, a novel multiscale template matching algorithm for multimodal image matching is proposed to address this problem. We propose a novel frequency-domain convolutional map based on the wavelet transform and phase congruency to construct a feature description map that significantly reduces the NRD between multimodal images. The development of omnidirectional aggregated feature vectors with rotational invariance also helped to achieve robustness on rotated images. Finally, a multiscale template matching strategy improved the matching performance on multimodal images with displacement and scale variations. To improve the time efficiency of the algorithm, most of the complex computations in this article are performed in the frequency domain. According to the experimental findings on six multimodal image datasets, the method can obtain accurate and robust matching results between multimodal images. Through qualitative and quantitative evaluations, the method outperforms several mainstream multimodal image matching algorithms in terms of matching accuracy, success rate, and time consumption.

Keywords

• Multimodal remote sensing images
• multiscale strategy
• template matching
• wavelet transform