In our data, the overlapping tile areas are significantly photobleached due to second acquisition exposure.
We observe a factor of 10 difference in the number of detected interestpoints (with the same intensity and threshold parameters) in the pariwise overlapping area between the tiles. This can disrupt the geometric descriptor based matching.
As interestpoint intensities can be available, we propose the introduction of --pairwiseBrightest=N option for the --interestpointsForReg=OVERLAPPING_ONLY case. If N>0, then in the pairwise matching step, the number of interestpoints should be limited to the min(N,N_ta,N_tb) brightest ones on each tile in the pairwise overlapping area. (N_ta, N_tb is the number of points in the tileA pairwise overlapping area and tileB pairwise overlapping area respectively.) So that the number of interestpoints becomes equal that go into the descriptor based matching and the brightest ones supposed to be the same on both tiles.
(Or maybe min(N, 2 N_ta, 2 N_tb) to allow for some noisy detections.)
In our data, the overlapping tile areas are significantly photobleached due to second acquisition exposure. We observe a factor of 10 difference in the number of detected interestpoints (with the same intensity and threshold parameters) in the pariwise overlapping area between the tiles. This can disrupt the geometric descriptor based matching.
As interestpoint intensities can be available, we propose the introduction of
--pairwiseBrightest=Noption for the--interestpointsForReg=OVERLAPPING_ONLYcase. If N>0, then in the pairwise matching step, the number of interestpoints should be limited to the min(N,N_ta,N_tb) brightest ones on each tile in the pairwise overlapping area. (N_ta, N_tb is the number of points in the tileA pairwise overlapping area and tileB pairwise overlapping area respectively.) So that the number of interestpoints becomes equal that go into the descriptor based matching and the brightest ones supposed to be the same on both tiles. (Or maybe min(N, 2 N_ta, 2 N_tb) to allow for some noisy detections.)