Algorithm 1.

Auto-Montaging with ORB

Input:    Tiles T = {t1, t2, . . ., tn} with ti = (ci, si, di) consisting of three images
Nominal positions P = {p1, p2, . . ., pn}
Output:    Transformations between images
Calculate ORB keypoints and descriptors of all tiles
Associate nominally close tiles
while Still an unaligned image do
Pick random unaligned tile and consider aligned
while aligned a tile on previous iteration do
for src ∈ unmatchedImages do
for dst ∈ nominallyClose(src) ∩ matchedImages do
Compare src and dst descriptors using LSH
Estimate transformation, and inliers using RANSAC, saving result
if #inliers > 50 then	▹ Translation is good enough
Save transformation from src to dst
Exit for loop
if best #inliers > 10 then	▹ Best translation is good enough
Save transformation from src to its best match