Face attribute refined editing method based on generative adversarial network hidden space deconstruction

Abstract

The invention discloses a face attribute refined editing method based on generative adversarial network hidden space deconstruction. The method comprises the following steps: S101, constructing a generator model; S102, constructing a classifier model; S103, modifying generation codes through a normal vector to obtain a pair of generation codes; S104, substituting the pair of generation codes intothe generator model to obtain a pair of face images; S105, substituting the pair of face images into the classifier model to obtain a binary classification result; S106, calculating the minimum valueof the loss function; S107, repeating the steps S103 to S106 through a reverse gradient algorithm, optimizing the normal vectors to move the generation codes in the positive and negative directions, enabling the face attributes to change, and taking the normal vector when the change degree of the face attributes is maximum as a target normal vector; and S108, generating a new face image with the face attribute corresponding to the target normal vector.

Description

technical field [0001] The invention relates to the technical fields of digital image processing and face image editing and synthesis, in particular to a refined editing method of face attributes based on hidden space deconstruction of generative adversarial networks. Background technique [0002] Face attribute editing refers to the process of modifying one or more attributes of a face image to generate a new face image with target attributes. The editing and generation technology of face images has important applications in the fields of public security, digital entertainment and other livelihood needs. In the field of public security, when a crime occurs when the surveillance is not covered or covered, law enforcement officers cannot directly obtain the suspect's face photo. Face portraits with specified attributes drawn based on eyewitness testimony have become the main way to identify suspects. In the field of digital entertainment, beauty styling design technology ac...

AI Technical Summary

Problems solved by technology

This type of method can add or remove attributes from face images to achieve qualitative modification of attributes, but it is still difficult to achieve quantitative modification of specified attributes

[0004] To sum up, the existing methods have considerable limitations, and can only transfer the attributes of one face to another face as a whole, or qualitatively modify the attributes of the face by adding or removing them.

Images