On the Performance of Multiwavelets for Handwritten Numeral Recognition

In this paper we study the recognition of contour-base hand written numeral characters using Multiwavelets and neural networks. For reaching this purpose, in the first step we extract numerals chain code by tracing their contours. Then we perform multiwavelet transform on it to prepare the appropriate features. Finally by employing the feed forward neural network we classify them into predefined classes. The neural network was trained with handwritten numeral database, MNIST. The experiments have demonstrated that the multiwavelet and neural network system is able to more correct recognition of digits of the MNIST test set.

Keywords:

Neural network , Multiwavelets , character recognition , Feature extraction , Shape representation