A Feature-Based Method for Evaluating the Quality of Heterogeneous Image Matching

Image matching is a critical area of research in computer vision, with widespread applications in scene matching, image retrieval, and autonomous driving. However, accurately evaluating image matching quality without ground truth data remains a significant challenge. This paper presents a novel image matching quality evaluation method based on the Enhanced Generalized Dual-Bootstrap Iterative Closest Point (E-GDB-ICP) algorithm. The method begins by using GDB features for robust image representation, followed by feature point selection and determination of the image’s orientation based on the principal direction of the feature points. Finally, a matching evaluation function is employed to score the image matching quality. The method is extensively validated across various image datasets, including heterogeneous remote sensing and building images, and compared with traditional image matching evaluation methods. Results show that the proposed method not only accurately assesses matching quality for these images but also demonstrates superior robustness and adaptability in complex scenarios. These findings confirm the effectiveness and reliability of the proposed method for evaluating image matching quality in the absence of ground truth data.