The adaptive network combined for data compression
Andreas Weingessel, Horst Bischof, Kurt Hornik · 1997
We present a method to adaptively combine multilayer perceptrons and vector quantization. We show that this combined network yields better results than the sequential application. 1 Introduction In this paper we deal with the combination of principal component analysis (PCA) networks and vector quantization. The idea to use these both methods together comes from applications in image compression tasks. PCA is a method to reduce the dimensionality of the input by projecting it to the so-called principal subspace. By this projection the information which is orthogonal to the principal subspace is lost, so reconstructing the original data yields some error. The original data can be recovered by knowing the (lower-dimensional) principal subspace projection and the error made by the PCA reconstruction. To increase the rate of compression, the error of the PCA reconstruction is stored in a quantized form. Up to now such a combination has been done in a sequential manner. This means that fir...