Human face cartoon generation method based on learning geometry and texture style migration

Abstract

The invention discloses a human face cartoon generation method based on learning geometry and texture style migration. The method comprises the following steps: a human face deformation graph is acquired based on a geometric deformation module; the texture migration module follows a manifold alignment style transfer hypothesis and migrates a new style based on a neural network optimization mode; for a local similar semantic region of the deformation graph and the cartoon, a texture migration module carries out constraint stylization to output a feature graph similar to a cartoon style graph; and an artistic cartoon augmented style data set with various styles is generated through the StyleGAN, and various cartoons are acquired in a hidden space interpolation mode, so that diversified face cartoon generation is realized. The method can exaggerate the face characteristics of people and obtain geometric deformation styles of different artists in a customized manner; geometric distortion and texture rendering are performed in a targeted manner, and a cartoon style graph with controllable interpolation is combined, so that the generated face cartoon image is more vivid and diversified.

Description

technical field [0001] The invention belongs to the field of computer applications, and in particular relates to a method for generating human face caricatures based on learning geometry and texture style transfer. Background technique [0002] Face caricature is an art form that expresses specific emotions and impresses people by exaggerating and emphasizing the characteristics of tasks. It has rich and flexible diversity and is deeply loved by the public. On the one hand, face caricatures can be depicted in different forms, such as stick figures, sketches, oil paintings, etc.; on the other hand, face caricatures can also express different emotions through different exaggerated styles. At the same time, comic creators also have their own artistic styles and ways of expression, which further increases the diversity of the art form of face comics. [0003] Face caricatures usually need to be created by some artists with professional skills, so often only a few celebrities su...

AI Technical Summary

Problems solved by technology

However, these methods can only generate a fixed shape and exaggerated style for the same input photo, which cannot meet people's needs for the diversity of comic styles.

Generally speaking, the automatic generation of face caricatures has the following difficulties: 1) Generating a face caricature from a face photo not only requires the conversion of the color style of the image, but also needs to be based on the characteristics of the input face, artistic creation style, etc. The shape of the image is exaggerated; 2) The art form of face caricature has rich diversity. Caricature creation based on the same photo will also have different creative methods, different expressions of emotions, and different styles of artists. Various styles are generated in the color and shape of comics; 3) In addition, for the texture style transfer process, the main limitation is that the generated comic images will be limited to the styles provided by the dataset

Images