Author(s)

M. Oravec & P. Podhradsky

Abstract

Backpropagation neural network used for compression of X-ray and MR images is considered. Hidden layer outputs of a trained network are appropriately fonned to two-dimensional data and visualised. The formed two-dimensional images are compressed using various discrete orthogonal transforms with the possibility to decorrelate intrablock, as well as interblock correlations. 1 Introduction Neural networks have a number of important properties that befit their use for signal processing applications (neural networks are distributed nonlinear devices, a neural network consists of a massively parallel processor that has the potential to be fault toler

Keywords