Jensen-Shannon centroid

Consider the set of categorical distributions (normalized histograms) as a mixture family in information geometry. Jensen-Shannon divergence between two categorical distributions then amounts to a corresponding Jensen divergence. Uses the Convex-ConCave procedure to locally optimize the JS centroid.

CCCP algorithm for the Jensen-Shannon centroid (PDF)
Some Java code implementing the Jensen-Shannon centroid CCCP
Some Python code implementing the Jensen-Shannon centroid CCCP
Lena histogram and Barbara.histogram obtained from Lena grayscale image and Barbara grayscale image

Reference

On a Generalization of the Jensen–Shannon Divergence and the Jensen–Shannon Centroid, Entropy, 2020 (arXiv:1912.00610)