A Robust Invariant Local Feature Matching Method for Changing Scenes

Table 7

The experimental contrast results.

Matching algorithm	Local feature	HEstimation	Precision	Recall
AGNN + Sinkhorn	SIFT	0.656	0.625	0.413
AGNN + Sinkhorn	ORB	0.692	0.677	0.579
Nearest neighbor	SIFT	0.643	0.514	0.349
Nearest neighbor	ORB	0.687	0.572	0.421
Nearest neighbor	Superpoint	0.651	0.594	0.445
Nearest neighbor	Des + NetVLAD	0.687	0.625	0.494
SuperGlue [19]		0.668	0.725	0.596
Our proposed method		0.784	0.807	0.634
Promotion		0.097	0.082	0.038