A Probabilistic Neural Network to Recognize Handwritten Digits using Boundary Descriptor Properties

Recognition of handwritten digits is a challenging task, because the writers may possibly write with dissimilar styles, sizes, width and shapes. A probabilistic neural network for recognizing handwritten digits is proposed here. Normalization of the digits of varying sizes is done for getting better boundary descriptor properties. The different boundary descriptor features extracted for recognition are compactness, eccentricity, equivalent diameter, extent and solidity. Classification of these features is done with probabilistic neural network. These features are tested on MNIST digit data set and observed good results.