Chinese font conversion and generation method based on small number of target fonts

Abstract

The invention provides a Chinese font conversion and generation method based on a small number of target fonts. The method comprises the following steps: (1) preparing an original font stroke data setand an original font-target font data set; (2) selecting training samples with relatively high representativeness by adopting a simple image dimension reduction and clustering algorithm; (3) trainingan end-to-end deep stroke discrete font transformation (DeepSDFT) neural network, and storing network parameters; and (4) automatically generating Chinese characters in the target fonts by using theDeepSDFT neural network to acquire remaining Chinese characters in the target fonts. According to the Chinese font conversion and generation method based on a small number of target fonts, no manual interaction is needed, and no network pre-training is needed; a small number of network parameters are needed; the method can be used for generating various target fonts; interpretability of an embedded feature space is realized; the Chinese font conversion and generation method is applicable to a font design assisted scene and is relatively high in practicability.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a method for converting and generating Chinese character fonts based on a small number of target fonts. Background technique [0002] Text fills every part of modern life. People need to read a lot of information transmitted by different media through text. Stylized text can often grab the reader's attention at the first time. However, the number of Chinese characters is huge. Taking the Chinese national standard font GB 18030-2000 as an example, there are 27,533 Chinese characters in total. Even if a professional font designer designs a complete set of Chinese fonts, it will take 2 to 3 years. less efficient. Automatic generation of Chinese character fonts is one of the methods to solve this problem. According to the existing small number of target font Chinese characters, the remaining Chinese characters with the target style can be automatically generated, which can ...

AI Technical Summary

Problems solved by technology

This method produces accurate results, but requires that the difference between the skeleton shapes of the two fonts should not be too large, and the complexity of the deformation algorithm is high

The third method is based on a deep convolutional network, which trains the network to learn the mapping from an original font to a target font. This method generally does not require human intervention and can quickly complete model training, but the generation effect of more stylized target fonts is unstable. And requires a large number of training samples, which is not suitable for practical applications

Images