The Application of a Generic Neural Network to Handwritten Digit Classification
Dylan Banarse, A.W.G. Duller · 2020
This chapter analyses the application of a self-organizing neural network to handwritten character recognition. It describes the application of the Pattern Recognition Architecture for Deformation Invariant Shape Encoding (PARADlSE) neural network to the recognition of handwritten characters from the National Institute of Standards and Technology database. The PARADISE network was designed to be a generic recognition system with the solution of many recognition problems requiring a minimum amount of application-specific knowledge to be provided. The PARADISE network has a three layer architecture: the feature extraction layer, the pattern detection layer and the classification layer. A number of parameters can be set to control the type of recognition performed by the network. In order to use the network for a new recognition application, the most important aspect is to determine the type of feature extraction that is best suited to the objects. Two methods of feature extraction were chosen to be tested, Gabor filters and oriented Gaussian filters.