Effective feature matching of high-resolution planetary orbiter images based on optimized image partitioning and rapid local correspondence

Feature matching of high-resolution planetary orbiter images (HRPOIs) plays a significant role in photogrammetric mapping and three-dimensional modeling, which is particularly challenging due to their narrow fields of view, similar surface textures, and large sizes. Neither directly using matching methods nor combining them with traditional strategies can fully overcome the matching challenges posed by HRPOIs. In this study, a feature matching scheme specifically designed for HRPOIs is proposed, which integrates an optimized image partitioning method for overlapping areas and establishes rapid local correspondence between image blocks. A row-to-column grid partitioning method is developed to ensure the range of the partitioned image blocks better align with the boundaries of the complex overlapping areas. Subsequently, the relative positional relationships of the extreme coordinate points in the row direction within the overlapping areas of image pairs are determined to rapidly establish local correspondence between the partitioned images, providing local geometric constraints and avoiding interference caused by redundant features. Moreover, a multilevel efficiency optimization strategy, which combines Graphics Processing Unit (GPU) acceleration with distributed parallelism to enhance the matching efficiency, is designed. Experiments on data derived from the Lunar Reconnaissance Orbiter Narrow Angle Camera (LRO NAC) and Mars Reconnaissance Orbiter Context Camera (MRO CTX) were conducted to validate the feasibility and reliability. The experimental results demonstrate that the proposed scheme effectively overcomes the challenges faced when matching HRPOIs, outperforming several mainstream photogrammetric software in terms of both matching accuracy and efficiency.