Adaptive sensor fusion using stochastic vector quantisers

A stochastic generalisation of the standard Linde-Buzo-Gray (LBG) approach to vector quantiser (VQ) design is presented, in which the encoder is implemented as the sampling of a vector of code indices from a probability distribution derived from the input vector, and the decoder is implemented as a superposition of reconstruction vectors. This stochastic VQ (SVQ) is optimised using a minimum mean Euclidean reconstruction distortion criterion, as in the LBG case. Numerical simulations with stereo pairs of images are used to demonstrate how this can lead to various types of self-organisation of the SVQ, each of which encodes and fuses the information in the stereo pair in a characteristic way.