Holistic word case recognition using a multi-layer perceptron neural network

The paper describes how a standard multi-layer perceptron (MLP) neural network can be used to correctly classify handwritten words according to whether they contain wholly upper-case or wholly lower-case characters. This without actually having to recognise any of the individual characters. Using an optimised 6-2-1 architecture MLP neural network, trained with the conventional backpropagation algorithm, it is shown that it is possible to successfully classify 84% of a 1061 word data set. This data set being randomly selected from a 3183 word data set obtained from 12 writers, each submitting approximately 150 words of both case