I don’t think this is a bug. It’s pretty typical for the algorithm to “see” different points of interest in a photo than a human eye would be naturally drawn to and find similarities that don’t appear to be all that similar to us. There’s actually an open feature request related to this: Better reference images in ranked match results