Image registration with semantic information-guided focus on co-pixel points

The goal of image registration is to align pixel points from images captured by different sensors, serving as a reliable foundation for subsequent information fusion and enhancing the performance of advanced vision tasks in degraded scenarios. However, current methods overlook the viewpoint variations caused by internal sensor differences, which introduce information of irrelevant interference, challenging the reliability of registration. Therefore, this paper proposes a semantic information-guided image registration algorithm that directs the network to focus on relevant pixel regions by leveraging the semantic information of corresponding pixels. Through comparative experiments, the proposed method demonstrates superior registration performance compared to existing methods.