Measure of Similarity between GMMs Based on Geometry-Aware Dimensionality Reduction

Gaussian Mixture Models (GMMs) are used in many traditional expert systems and modern artificial intelligence tasks such as automatic speech recognition, image recognition retrieval, pattern speaker verification, financial forecasting applications others, simple statistical representations of underlying data. Those typically require high-dimensional GMM components that consume large computing resources increase computation time. On the other hand, real-time computationally efficient algorithms for reason, various similarity measures dimensionality reduction techniques have been examined to reduce computational complexity. In this paper, a novel measure is proposed. The based on recently presented nonlinear geometry-aware algorithm manifold Symmetric Positive Definite (SPD) matrices. applied over SPD original local neighborhood information from parameter space preserved by preserving distance mean. Instead dealing with space, method operates much lower-dimensional transformed parameters. Resolving between reduced calculating among was tested within texture task where superior state-of-the-art performance terms trade-off accuracy complexity has achieved comparison all baseline measures.