Riemannian \(L^p\) averaging on Lie group of nonzero quaternions (Q742343)

The author studies two families of algorithms for the computation of the weighted mean of quaternions in the intrinsic geometry of \(\mathbb{H}^*\) endowed with its natural bi-invariant Riemannian metric. Some algorithms for computing the Riemannian \(L^p\) center of mass of a set of points, with \(1\leq p\leq \infty\) (median, mean, \(L^p\)-barycenter and minimax center) are particularized to the case of \(\mathbb{H}^*\). The author uses two approaches. The first one is based on computing the logarithm of quaternions (the map logarithm map sends the quaternions to the Euclidean tangent space of \(\mathbb{H}^*\) at the identity. The second one is a family of methods based on the gradient descent algorithms aiming at minimizing the sum of quaternion distances raised to the power \(p\). After a short description of the multiplicative Lie group of nonzero quaternions, the author presents some aspects from the Riemannian geometry structure of \(\mathbb{H}^*\). Then he presents the averaging on the quaternion Lie group \(\mathbb{H}^*\) by defining the Riemannian \(L^p\) center of mass and log-Euclidean \(L^p\) averaging on \(H^*\) (log-Euclidean mean, log-Euclidean median, log-Euclidean minimax center and properties of log-Euclidean \(L^p\) averages). Next he studies the Riemannian \(L^p\) averaging in \(\mathbb{H}^*\) (Riemannian mean, Riemannian median, Riemannian \(L^p\) center of mass, Riemannian minimax center). Finally, the author studies the bilateral filtering of quaternion images (rational, blurring effect in Riemannian \(L^p\) filtering, locally adaptive Riemannian \(L^p\) averaging).