Closed ieivanov closed 3 weeks ago
This version uses an average row_translation
and column_translation
to stitch all images in a grid. The average row_translation
and column_translation
are computed using phase cross correlation on a small number of tiles (e.g. 3x3) or using a calibration dataset. No registration is performed between tiles during stitching of larger datasets. It works decently well, but I've seen examples where the stitch is not good in certain sections of the mosaic, likely because the XY stage makes a larger error when moving.
I'm planning a V2 which will compute shifts between all pairs of neighboring images using phase cross correlation. This can be done quickly on the HPC. All images in a given row share the same row_translation
- we can compute an average row_translation
for each row. Each pair of images will have a unique column_translation
. The phase cross correlation algorithm could make errors in estimating the shift - we'll replace outliers with the mean value across the dataset.
Replaced by #137
TODO:
compute_image_translation
to CLI and save in config file