Comparison of 3d curves on faces collected from media pipe landmarks

Question

I'm working on a project where I am trying to perform facial recognition by comparing facial features mapped with 3D curves captured on faces using ~300 media pipe landmarks. The idea is that media pipe collectes the 3D positions, curves which I have pre-defined along those landmarks extracted from face A and B and then a comparison is performed to determine how similar they are. (for now I am only using head on faces, at a clear resolution and with mostly neutral expressions) I'm at the stage of comparing the facial features and there doesn't seem to be much correlation in my results, that is to say that the comparison's are not finding the positive matches to have much more similairty than the negative.

So far for the comparison I have tried using MSE, an angle based approach, DTW (very briefly) and most recently procrustes analysis, they seem to yield similar results however they are not correctly matching identities or returning particularly high similarity for matching identities (for clarity at the moment my approach produces a score which indicates the total difference between faces using procrustes analysis, previously it was MSE, when I say high similarity I mean low difference), I planned to next go through the contributions of each of my defined facial features and analyse their normalised contributions to false negatives (by dividng the difference score they contribute by the number of landmarks they contain) to try and remove any features which were stronly influencing these false negatives but wanted to see if there were any glaring issues with my approach first. I am aware that this is by no means the best method of facial recognition, the project is more about the geometric approach than the actual accuracy, thank you in advance.