Single layey neural network, 40*40 pixels(1600 dimensions), 27 words dataset:
1600 inputs---->27 neurons(using sigmoid)---->27outputs
Implemented base on https://github.com/timmmGZ/MachineLearning-Perceptron
I haven't draw too much into the dataset, if the probabilty is less than 90% when you draw some word correctly, please add the drawing to train set, then train around 2000 times again.
Here has a little bit update about generating random drawing:
.JPG)