I’ve implemented the Iterative Closest Point algorithm in G’MIC, with a new command match_icp.
I think you may be interested by this one, as it provides another way to make two sets of N-D vectors match together, by applying a linear transformation.
Here is an example to register two 2D shapes:
foo :
# Create two distincts silhouettes to match.
shape_cupid 400 shape_cupid 300 rotate. 65 deform. 5 gt. 50%
# Convert silhouette contours as set of 2D points.
foreach {
(0,1,0;1,1,1;0,1,0) +erode.. . rm.. -
1,1,1,2 eval.. ">i?da_push(#-1,[x,y])" k. da_freeze
}
# Match the two sets of points.
+match_icp. ..
# Draw result.
400,400,1,3
eval[0] "I(#-1,i0,i1) = [ 255,0,0 ]" # Red: reference contour
eval[1] "I(#-1,i0,i1) = [ 0,0,255 ]" # Blue: contour we want to match the reference
eval[2] "I(#-1,i0,i1) = [ 0,255,0 ]" # Green: blue contour linearly transformed to match red contour
k.
Result:
I had the idea of using this for doing color transfer, but so far, it doesn’t work that well.
Anyway, still cool to have this integrated into G’MIC !
