I'm working on a problem in image processing where I am looking to group detected curves. Basically I'm looking for a good metric of the similarity between a pair of quadratic curves in a discrete real space. Then if the two curves are close enough (basically the same curve) then I can refit a new curve using the combination of data used to fit the two curves separately.
The curves are fairly smooth and generally almost linear with maybe a slight gradual curvature. No sharp turns.
L2 on their individual coefficients feels wrong and I'm wondering if there is a better method than say average distance of sampled points or something.