Use simple Euclid-norm to calculate the distance between 2 images when the size are small. When it comes to larger size images, use recur_align to align it with the help of image pyrimad, and adapt a more efficient metric to measure the distance, which is ncc, distance after normolization.
Some results of the example data.
Some results of the data in Prokudin-Gorskii collection(my favorite pictures!).
Implemented a auto_crop function in align.py to crop the edges of the images, but seems not help that much.