Vector quantization for image classification with side information for the additive Gaussian noise channels

Gauss mixture vector quantizers (GMVQ's), designed using the Lloyd algorithm, provide an approach to the image classification problems, utilizing the robustness and the analytical tractability of the Gaussian distribution. We generalize the Lloyd-based GMVQ training algorithm to design a Lloyd-optimal GMVQ when only a noisy version of the original data is available at the classifier and the classifier is allowed to cooperate with sensors, having different noisy versions of the original data, under rate constraints. Our simulations, using a set of aerial images, indicate that our algorithm leads to a better classification performance than the non-optimized schemes.