出版社：International Association for Computer Information Systems
摘要：A robust algorithm to train neural network (NN) with individual character presented in pixel format is described. The output of the trained NN is related with the pixel binary format. The trained NN performs the classification based on the original pixel format code presented to the NN. A mapping scheme is proposed to determine the self-adaptive variable learning rates for rapid converge. The proposed character recognition method trains at a faster rate than that of the standard back propagation training method and prevents oscillations during training.
关键词：Self-adaptive training; ANN; Dynamic variable learning rates; Character recognition.
