Image compression using feedforward neural networks - Hierarchical approach

Abstract

The paper presents the hierarchical approach to the problem of image compression using feedforward neural networks. In this approach smaller frames are used in the regions containing more details and larger, when the grey level is uniform. Thanks to this the number of data is reduced, learning speed accelerated and the quality of compression improved. The numerical results, confirming the efficiency of the proposed approach and good generalization properties are presented and discussed.