A robust method for still image compression using dynamically constructive neural network

Abstract

A dynamically constructive neural network (DCNN) is proposed for still image compression. The main feature of the proposed dynamical construction is its robustness to input-to-hidden and hidden-to-output link failure. A wavelet transform based sub-image block classification technique is also proposed for partitioning training images into image clusters. Each cluster is used as a training set for training a particular DCNN. This ensures the generalization capability of DCNNs. Computer simulation results demonstrate superiority of the proposed scheme in terms of peak signal to noise ratio and robustness as compared to that of other recent methods.