Homologous Image Registration Algorithm Based on Generalized Dual-Bootstrap Iterative Closest Point Algorithm

Heterogeneous image registration is a critical field in computer vision, with broad applications in scene registration, image retrieval, and autonomous driving. Due to differences in sensor types and imaging conditions, heterogeneous images often exhibit significant appearance discrepancies, leading to errors when using traditional registration methods. To address this challenge, we propose the Enhanced Generalized Dual-Bootstrap Iterative Closest Point (E-GDB-ICP) algorithm, designed to improve the accuracy and robustness of heterogeneous image registration. The proposed algorithm first utilizes SIFT descriptors and a multi-scale keypoint detector to initialize feature points, significantly increasing initial registration accuracy. The feature point selection process is then refined to further enhance the reliability of matches. Additionally, a new image registration quality evaluation function is introduced, enabling the assessment of registration outcomes without requiring ground truth data. This makes it a valuable tool for real-world applications. Experimental validation on remote sensing and architectural image datasets demonstrates the effectiveness of the proposed method, which achieves improved accuracy and robustness in heterogeneous image registration tasks.