A graph matching based key point correspondence method for lunar surface images

Key point correspondence plays an important role in lunar surface image processing. Since lunar surface images often contain obvious illumination changes, noisy points and repetitive patterns, traditional appearance based algorithms may fail when local appearance descriptors become less distinctive. In this paper, we introduce a graph matching based algorithm to tackle this problem. First, by incorporating structural information, key point sets in lunar surface images are represented by graphs. Then key point correspondence is formulated as a specific graph matching problem which aims to find a specified number of best assignments, and effectively approximately solved. Finally, an outlier assignment elimination method is proposed based on the affine invariance assumption. Simulations on both benchmark datasets and lunar surface images witness the effectiveness of the proposed method.