Three-dimensional face modeling method based on double-tributary network

Abstract

The invention discloses a three-dimensional face modeling method based on a double-tributary network, and the method comprises the following steps: 1) constructing a face shape reconstruction tributary based on an auto-encoder structure, and obtaining an identity space diagram and an expression space diagram; 2) respectively expressing an identity attribute component and an expression attribute component by utilizing the identity space diagram and the expression space diagram; 3) constructing a facial texture reconstruction branch based on a semantic region to obtain a facial texture map withhigh truth; and 4) constructing a fusion module: correspondingly combining the reconstructed face shape model and the reconstructed face texture map according to a predefined three-dimensional vertextopological relation to obtain a final reconstructed three-dimensional face model. According to the invention, decoupling learning is carried out on the identity and expression attributes of the humanface, and processing of the face shape and the face texture is carried out by using the double-tributary network, so that accurate three-dimensional human face reconstruction with large-amplitude expressions is achieved.

Description

technical field [0001] The invention belongs to the fields of computer vision and computer graphics, and in particular relates to a three-dimensional human face modeling method based on a double-branch network. Background technique [0002] Monocular face reconstruction aims to recover the corresponding 3D face model from a single face image. In recent years, because of its wide application in practice, such as face alignment, face editing, and virtual reality, a lot of research work has been invested in this field. [0003] However, it is very challenging to reconstruct accurate face geometry and recover realistic facial texture maps from a single image. One of the inescapable challenges is the diversity of facial expressions, an innate human property. The diversity and ambiguity of facial expressions have become the key problems that must be solved in the process of 3D face reconstruction. [0004] In order to improve the effect of face reconstruction, many methods have...

AI Technical Summary

Problems solved by technology

However, the amount of data available for training facial texture maps is very limited, and there are disadvantages of difficulty in obtaining and high cost, so this method limits its application range

[0005] To sum up, the existing methods either use the linear 3DMM method to estimate the model parameters with limited expression ability by using the neural network; or use the unconstrained three-dimensional face expression method, but do not specifically deal with the attributes of the facial expression. , so it will cause prediction errors under the condition of large-scale expressions, thereby reducing the final 3D face reconstruction accuracy

Images