Automatic Pattern Recognition with Wavelet Neural Network

Abstract

Automatic recognition of prototype is very broad field which find more and more application in different areas. It consists to categorize and recognize manuscript, letters, or numbers. Several techniques have been applied to deal with the recognition of prototype. Inspired by the operation of the nervous system, we propose, in this paper, to build up artificial neural network of type Wavelet-Multi-Layer Perceptrons (WMLPs) to deal with recognition of numbers captured using a flatbed scanner. We focus on the use of wavelet transfer function and their importance in neural network approach. In addition, we show during this study the essentials stages to go through in order to be able to deal with recognition of number by WMLPs. The data used to train the WMLPs are 63 in total; nevertheless, the result confirmed the accuracy and the effectiveness of WMLPs in the field of number recognition.