Recognition on Matrix Angular Central Gaussian Distribution

We demonstrate the standard approach of Maximum Likelihood Estimation (MLE) for practicability of Grassmann Angular Central Gaussian (GACG) distribution by using Grassmann manifold. Our main concern is then on the applicability of GACG for computer vision application e.g., classification on arbitrarily high dimensional Grassmannian space. We show by numerical experiments that the implementation of the proposed Grassmannian variate parametric model via MLE using simple Bayesian classifier is directly related to the accurate calculation of normalising constant naturally appearing with them. We verify the validity and performance of our proposed approach on two publicly available databases against the existing state of art techniques, where we observed that the classification accuracy of our proposed approach outperforms significantly.