Optimal Wavelets and Neural Networks for Pattern Recognition
Guoliang Chen, Tien Dai Bui, Adam Krzyżak · 2003
We investigate the applications of optimal wavelets and neural networks in the recognition of handwritten numerals. Wavelet transforms have been successfully applied in many applications including pattern recognition. However, which kind of wavelet should be used is still an open problem. We propose to use a combination of optimal wavelets and neural networks in pattern recognition applications. The optimal wavelet filters can be obtained by minimizing the mean square error of the neural network output. The same error objective function is used for training the weights of the neural network. We conducted some experiments on unconstrained handwritten numeral recognition and observed 1 60% increase in the recognition rate compared to the Daubechies-4 wavelet on the Concordia handwritten numeral database.