In this work, an analytical model has been developed to demonstrate classification performance when fusing two quantized features. Specifically, it is of interest to demonstrate theoretically the effect that the overall quantization of the features, M, has on the relative performance of the Bayesian Data Reduction Algorithm (BDRA). The primary results show, and with a training data model independent of distribution, conditions on the data under which dimensionality reduction improves overall theoretical classification performance. This result is significant for those interested in the theoretical performance of fusing discrete data (i.e., attributes or classifier decisions), and is an important step towards proving that BDRA always converges to a unique solution.

Date Published: 22 May 2015
PDF: 10 pages
Proc. SPIE 9498, Multisensor, Multisource Information Fusion: Architectures, Algorithms, and Applications 2015, 949806 (22 May 2015); doi: 10.1117/12.2180063

Published in SPIE Proceedings Vol. 9498:
Multisensor, Multisource Information Fusion: Architectures, Algorithms, and Applications 2015
Jerome J. Braun, Editor(s)