Results of data compression for plane curves using neural networks

This paper provides some results of comparing two neural net structures' performances in compressing input patterns from a bidimensional original curve. We apply an interpolation algorithm based on nonlinear fuzzy rules to regenerate the compressed information. Resulting interpolation errors and compression capacity are features that are analysed. As an illustration we apply the compression system and interpolation method to some curves, and show identical and different results for two models-symbolic and numerical. In terms of abstraction capacity, better results were obtained for symbolic processing model. The excellent results demonstrate the capability of compression system in extracting the minimal necessary information.<
>