Abstract
A high-order feedforward neural architecture, called pi t -sigma (π t σ) neural network, is proposed for lossy digital image compression and reconstruction problems. The π t σ network architecture is composed of an input layer, a single hidden layer, and an output layer. The hidden layer is composed of classical additive neurons, whereas the output layer is composed of translated multiplicative neurons (π t -neurons). A two-stage learning algorithm is proposed to adjust the parameters of the π t σ network: first, a genetic algorithm (GA) is used to avoid premature convergence to poor local minima; in the second stage, a conjugate gradient method is used to fine-tune the solution found by GA. Experiments using the Standard Image Database and infrared satellite images show that the proposed π t σ network performs better than classical multilayer perceptron, improving the reconstruction precision (measured by the mean squared error) in about 56%, on average.
Similar content being viewed by others
References
Duch W, Jankowski N (1999) Survey of neural transfer functions. Neural Comput Surv 2:163–212
Fahlman SE, Lebiere C (1991) The cascade-correlation learning architecture. Technical Report CMU-CS-90–100, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA
Ghosh J, Shin Y (1992) Efficient higher-order neural networks for classification and function approximation. Int J Neural Syst 3:323–350
Haykin S (1999) Neural networks – a comprehensive foundation 2nd edn. Prentice Hall, Englewood Cliffs
Iyoda EM, Nobuhara H, Kawamoto K, Yoshida S, Hirota, K (2003a) Sigma-pi t Cascaded hybrid neural network and its application to the spirals and sonar pattern classification problems. In: Proceedings of the 4th international symposium on advanced intelligent systems, Jeju Island, Korea, pp 158–161
Iyoda EM, Nobuhara H, Hirota K (2003b) Solving the N-bit parity problem using a single translated multiplicative neuron. Neural Process Lett 18:213–218
Iyoda EM, Nobuhara H, Hirota K (2004) Translated multiplicative neuron: an extended multiplicative neuron that can translate decision surfaces. J Adv Comput Intell Intell Inf (to appear)
Jiang J (1999) Image compression with neural networks – a survey. Signal Process Imag Commun 14:737–760
Leerink LR, Lee Giles C, Horne BG, Jabri MA (1995) Product unit learning. Technical Report CS-TR-3503, University of Maryland, College Park, MD
Ma L, Khorasani K (2002) Application of adaptive constructive neural network to image compression. IEEE Trans Neural Netw 13:1112–1126
Michalewicz Z (1996) Genetic algorithms + data structures = evolution programs, 3rd edn. Springer, Berlin Heidelberg New York
Møller MF (1993) Scaled conjugate gradient algorithm for fast supervised learning. Neural Netw 6:525–533
Namphol A, Chin SH, Arozullah M (1996) Image compression with a hierarchical neural network. IEEE Trans Aerosp Electr Sys 32:326–337
Schmitt M (2002) On the complexity of computing and learning with multiplicative neurons. Neural Comput 14:241–301
Zhang B-T (2002) A bayesian evolutionary approach to the design and learning of heterogeneous neural trees. Integr Comput Aided Eng 9:73–86
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Iyoda, E.M., Shibata, T., Nobuhara, H. et al. Image Compression and Reconstruction Using pi t -Sigma Neural Networks. Soft Comput 11, 53–61 (2007). https://doi.org/10.1007/s00500-006-0052-z
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00500-006-0052-z