Generative adversarial network face correction method and system based on identity constraint

Abstract

The invention discloses a face correction method and system of a generative adversarial network based on identity constraints. The generative adversarial network based on the identity constraints is constructed and trained. The network comprises a generator, a discriminator and a face feature extraction network. The generator comprises an encoder, an angle posture classification module, an identity constraint recognition module and a decoder. By designing an angle posture classification module and an identity constraint recognition module,the posture coding features and identity coding features are respectively acquired, and a feature loss function is introduced to constrain the identity coding features of the side face to approach the identity coding features of the front face; then identity constraints are added to the side face coding features, and decoupling of identity information and posture information in the side face coding features is realized, so that the side face identity coding features can generate a front face image with identity consistency through a decoder.

Description

technical field [0001] The invention belongs to the technical field of image generation, and in particular relates to a face correction method and system based on an identity constraint-based generative confrontation network. Background technique [0002] Face recognition is essentially a passive biometric technology for identifying uncooperative objects. In the real unconstrained environment, the accuracy of face recognition is greatly reduced due to factors such as posture changes, lighting, expressions, and occlusions. Although the deep convolutional neural network can represent image features very strongly, it turns out that a series of problems above have a great impact on the final performance of face recognition, especially when there are large changes in pose, especially When the face deflection angle is close to 90°, the accuracy of face recognition drops rapidly at this time. Face correction and re-recognition can currently improve the accuracy of face recognition...

AI Technical Summary

Problems solved by technology

In the past, the general process of face correction based on the generative confrontation network was that the side face image obtained the encoding features through the encoder in the generator, and then the front face image was generated through the decoder, and image discrimination and identity discrimination were performed on the generated front face image. To ensure that the generator generates a front face image that is as realistic as possible while maintaining identity information consistency, but does not add identity information constraints in the process of generating a front face image from a side face image, which cannot ensure that the generated front face image is consistent with Consistency of real frontal images

Images