A compression method for color images using multilayered neural networks

Multilayered neural networks can obtain compressed representations for input patterns in the hidden layer, when output patterns are identical with input patterns. However, when color images have complex features, a single neural network model cannot compress them with sufficient accuracy. In this paper, a compression method for color images is proposed based on a group of multilayered neural networks, which can reduce the reconstructed error in stepwise fashion. Finally, the compression and decoding performances of the proposed method are discussed in detail. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(2): 13–22, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10208