Evolving Stencils for Typefaces: Combining Machine Learning, User’s Preferences and Novelty

Table 1

Behaviour features.

feature	description
recognisability	character recognition using a Convolutional Neural Network (confidence value of the classifier in recognising the potential glyph as the target character)
similarity	visual similarity to existing glyphs using Self-Organising Maps (RMSE pixel-by-pixel similarity between the potential glyph and the most similar neuron in the Self-Organising Map of the target character)