For some project in computer vision I have N points in high-dimensional space. I want to select k of them that will be "the most distinguishable" from each other. For example, it can translate to sum of distances between chosen points is maximum. Or it can be that volume of polyhedron is maximum. But generally anything that has some intuition behind can go.
As expected I want to find these representative points.
There are two questions:
- What definition for "the most distinguishable" points is more commonly used? Do they change the algorithm used to find those points?
- What is the algorithm to find the points? It highly reminds me maximal weighted clique problem. Is it NP-hard problem? In this case can we make some good approximation against optimal solution?