Title :
A hybrid NN-Bayesian architecture for information fusion
Author :
Pan, H. ; Liang, Z.-P. ; Anastasio, T.J. ; Huang, T.S.
Author_Institution :
Beckman Inst. for Adv. Sci. & Technol., Illinois Univ., Urbana, IL, USA
Abstract :
This paper discusses a novel technique for information fusion. Specifically, a formula is derived for estimation of the joint probabilities in the maximum entropy sense. In addition, neural networks are used to estimate conditional probabilities required in the Bayesian inference method. Preliminary experimental results demonstrate that the proposed method can significantly improve the accuracy of the bimodal recognition system using audio/video signals
Keywords :
Bayes methods; audio signal processing; inference mechanisms; neural net architecture; probability; sensor fusion; speech recognition; video signal processing; Bayesian inference method; audio/video signals; bimodal recognition system accuracy; conditional probabilities; experimental results; hybrid NN-Bayesian architecture; information fusion; joint probabilities estimation; maximum entropy; neural networks; speech recognition; Bayesian methods; Computer architecture; Defense industry; Entropy; Military computing; Neural networks; Probability; Sensor fusion; Sensor phenomena and characterization; Training data;
Conference_Titel :
Image Processing, 1998. ICIP 98. Proceedings. 1998 International Conference on
Print_ISBN :
0-8186-8821-1
DOI :
10.1109/ICIP.1998.723502
