Fast adaptive algorithms for optimal feature extraction from Gaussian data

Abstract

We present a new adaptive algorithm to accelerate optimal feature extraction from a sequence of multi-class Gaussian data in order to classify them based on the Bayes decision rule. The optimal Gaussian feature extraction, in the Bayes sense, involves estimation of the square root of the inverse of the covariance matrix, Σ−1/2. We use an appropriate cost function to find the optimal step size in each iteration, in order to accelerate the convergence rate of the previously proposed algorithm for adaptive estimation of Σ−1/2. The performance of the proposed accelerated algorithm is compared with other adaptive Σ−1/2 algorithms. The proposed algorithm is tested for Gaussian feature extraction from three classes of three-dimensional Gaussian data. Simulation results confirm the effectiveness of the proposed algorithm for adaptive optimal feature extraction from a sequence of Gaussian data.