Optimizing wavelet transform coding using a neural network

Artificial neural nets have become very popular in various applications like pattern recognition, system identification and adaptive control. In general, a neural net is a nonlinear mapping device for function approximation in such a way that the arising error has to be minimized. This leads to an optimization problem where the cost function can be learned instead of being represented by a theoretical model. For that reason we have to discover the essential characteristics of the function to find its representation with minimal redundancy. Considering data compression there are similar requirements which have to be met. In this perspective, feature extraction as a basic capability of neural nets has to be performed for optimizing data compression. We consider wavelet transform coding because of its good fitting properties in both the time and frequency domains. Therefore, the choice of the wavelet is significant for a good fit of the signal and leads to an optimization problem which can be solved by a neural net.