Multi-style font generation method of variational auto-encoder based on vector quantization

Abstract

The invention discloses a multi-style font generation method of a variational auto-encoder based on vector quantization. The multi-style font generation method comprises the following steps: establishing a sampling sample of a Chinese character picture with a fixed font style; depending on a sampled sample, cutting characters of a sampling sample in sequence through the debugged frame with the fixed size and the step length; wherein only one word is included in the debugged frame, reading out data in the frame, converting the data into a picture, storing the picture, namely cutting out an image of a single font, and randomly selecting the image as a data set of style fonts needing to be expanded after cutting to serve as final training data; inputting the obtained final training data intoa network structure for iteration and training, and inputting a loss value obtained by network output into an optimizer to update an encoder weight and a decoder weight, and quantize a parameter valueof an implicit vector; and substituting a single training sample and a to-be-expanded Chinese character data sample into the updated encoder and decoder for reconstruction to obtain a font sample after style expansion.

Description

technical field [0001] The invention belongs to the technical field of image processing and artificial intelligence deep learning, and relates to a multi-style font generation method based on a vector quantization variational autoencoder. Background technique [0002] As the carrier of information, text not only conveys the content that people want to express, but also the style of text is gradually developed by people as an art form. In terms of poster design, slogan design, etc., using appropriate fonts can attract readers' attention when reading, and also greatly improve the overall aesthetics of the text. Some existing office software and image processing software have built-in many commonly used style fonts, and there are also many other styles of fonts designed by individuals or companies on the Internet. However, the font library of some style fonts did not include some uncommonly used Chinese characters in the design, and users will have inconsistent font styles due...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a multi-style font generation method based on a variational autoencoder based on vector quantization, which solves the problem that some uncommon Chinese characters in the style font font library existing in the prior art do not have the same style, and other style fonts are designed The problem of time-consuming and labor-intensive font library

Images