Single-image three-dimensional face reconstruction method and system based on convolutional neural network

Abstract

The invention provides a convolutional neural network-based single-image three-dimensional face reconstruction method and system. The method comprises the steps: training an improved convolutional neural network model through a training sample set; obtaining a two-dimensional single image of a to-be-reconstructed three-dimensional face, and inputting the two-dimensional single image into the trained improved convolutional neural network model to predict and obtain a face three-dimensional deformation parameter; and obtaining a reconstructed three-dimensional face based on the three-dimensional deformation model according to the three-dimensional deformation parameters and the shape vector and the texture vector of the face in the given two-dimensional image. A VGG-16 network model is improved, and a batch normalization layer is added after output of each convolutional layer and full connection layer, so that the value output in the middle of each layer of the whole neural network is more stable, and finally, the output of the whole network is more stable. And drastic change of output close to an output layer caused by updating of model parameters in the training process is avoided. Through experiments, the stability and fidelity of three-dimensional face reconstruction by the method are verified.

Description

technical field [0001] The invention belongs to the field of image processing, and in particular relates to a three-dimensional face reconstruction method and system for a single image based on a convolutional neural network. Background technique [0002] In recent years, 3D face reconstruction has become a hot topic in computer vision, image recognition and other research fields. 3D face reconstruction technology can be divided into reconstruction based on multiple images from different perspectives and 3D face reconstruction based on a single image. In many situations in real life, there is often only one face image available. Therefore, 3D face reconstruction based on a single image has attracted the attention of domestic scholars. [0003] At present, scholars at home and abroad have proposed a variety of methods for 3D face reconstruction from a single image. For example, traditional methods include model-based methods and methods based on light and dark shape restorat...

AI Technical Summary

Problems solved by technology

[0004] Under unconstrained conditions, there are great differences in facial expressions, poses, textures, and internal geometry. The above three-dimensional face reconstruction methods are still not stable enough. The reconstruction results show that the face is incomplete, the reconstructed face is biased towards the average face shape, and the expression is not realistic enough. And other issues

Images