Real-time fully automatic high-quality 3D face reconstruction method based on a single face image

Abstract

The invention discloses a real-time automatic high-quality three-dimensional face reconstruction method based on a single face image. The method comprises steps that fitting of a parameterization model coefficient of an initial face image data set and deviation of each pixel point of a face area along the depth direction is carried out to acquire a face image data set T; after an expression and attitude coefficient of the set T is changed, a face image is rendered to acquire a smooth face image data set C; after deviation of the set T is changed, the face image is rendered to acquire a face image data set F with details; a depth neural network model CoarseNet is trained through utilizing the set C, a depth neural network model FineNet is trained through utilizing the set F; an input single face image is processed through utilizing the trained depth neural network model CoarseNet and the trained depth neural network model FineNet to acquire the parameterization model coefficient and deviation of the pixel points along the depth direction, and a three-dimensional face is reconstructed. The method is advantaged in that the complex time consuming optimization process can be avoided, and the high quality three-dimensional face image can be rapidly reconstructed.

Description

technical field [0001] The invention relates to the technical field of face image processing, in particular to a real-time fully automatic high-quality three-dimensional face reconstruction method based on a single face image. Background technique [0002] In recent years, with the rapid development of computer technology and communication technology, images, as a convenient way of information transmission, have become an important means for people to obtain information. Since face images are widely used in the fields of identity verification, human-computer interaction, monitoring, forensics, medical diagnosis and treatment, the processing and recognition of face images has become the main topic of machine recognition and computer vision. Furthermore, the human face is the most representative part of the body, and people can obtain extremely rich meanings based on facial information alone, such as a person's gender, race, age, health, emotion, and even occupation. [0003]...

AI Technical Summary

Problems solved by technology

However, these methods have some problems: the optimization based on key point information only uses sparse key point information, and the accuracy of 3D reconstruction is low; the color-based optimization requires time-consuming calculations and is more sensitive to lighting

Images