Abstract: In this article, we study the regularization of quadratic energies that are integrated over discrete domains. This is a fairly general setting, often found in, but not limited to, geometry processing. The standard Tikhonov regularization is widely used such that, for instance, a low-pass filter enforces smoothness of the solution. This approach, however, is independent of the energy and the concrete problem, which leads to artifacts in various applications. Instead, we propose a regularization that enforces a low variation of the energy and is problem specific by construction. Essentially, this approach corresponds to minimization with respect to a different norm. Our construction is generic and can be plugged into any quadratic energy minimization, is simple to implement, and has no significant runtime overhead. We demonstrate this for a number of typical problems and discuss the expected benefits.
0 Replies
Loading