Recognizing pale letters and color letters with multi-layered neural networks
M. Kawada, Akira Taguchi, N. Iijima, Hideo Mitsui, M. Sone · 2002
Neural networks (NNs) on backpropagation (BP) are capable of carrying out the required mapping because NNs can absorb vagueness of input patterns. In this paper, a system is proposed that can recognize pale letters and color letters without the preprocessor. The basic NNs used have three layers. However the method to increase the number of layers is proposed. We construct a multi(4,5)-layered NN with four and five layers on BP. As a result, increasing the number of layers makes the limit of recognizable density lower. The 4,5-layered NN can recognize almost white color letters written on white paper. It has the ability to recognize multi-valued data like uneven density letters and color letters.>