Machine Learning-Based Font Recognition and Substitution Method for Electronic Publishing.

This study proposes a font recognition algorithm based on a deep convolution neural network and a font substitution algorithm based on texture and grayscale features. The experiments show that the proposed font recognition method can effectively extract font features with a high recognition rate, without the prior knowledge of the text content and with good versatility. The substitution effect of the proposed font replacement method can better satisfy the subjective visual perception of the human eyes and easily expand. The research results can be used to improve the publication quality; ensure the best presentation effect when presented in different platforms; facilitate font retrieval and effectively protect font copyright.