A method of multi-view face 3D model reconstruction based on deep learning

Abstract

The invention discloses a multi-view face three-dimensional model reconstruction method based on deep learning, which belongs to the field of computer vision. The method includes: multi-illumination and multi-viewpoint virtual face image generation; face front view depth map generation; multiple independent and parallel convolutional neural network training; neural network training for weight distribution of each viewing angle; restoring the depth map output by the network Create a 3D mesh model of the face and perform vertex coloring. The method of the present invention restores the depth map through independent training of multi-viewpoint images, and then trains the weight distribution map of each view point and then performs deep fusion, thereby improving the accuracy of the three-dimensional reconstruction model of the human face under the premise of ensuring efficiency.

Description

Technical field [0001] The present invention relates to the field of computer visual, and more particularly to a method of reconstruction based on deep learning multi-view face three-dimensional model. Background technique [0002] The application of human face three-dimensional model in the fields of safety certification, film and television animation, medical science, etc. are very wide. However, it is very expensive to obtain real and detailed face information, such as using a three-dimensional laser scanner. And multi-view is based on deep learning, and there is a fast, low cost, etc. The proposed image-based face three-dimensional reconstruction algorithm can be divided into two major categories: [0003] The first type of method is a three-dimensional reconstruction based on multi-view. Usually uses a traditional method, first acquire a faceful view, then use the motucture structure algorithm for the camera parameters, and then perform stereo matching, output the depth map ...

AI Technical Summary

Problems solved by technology

[0006] The above-mentioned prior art has at least the following disadvantages: the selected training data of almost all methods based on deep learning is based on the three-dimensional deformation model proposed by Blanz and Vetter, and usually the 3DMM face parameters are used as the input of the neural network for network training. Predict the 3DMM face parameters corresponding to the input image and restore the 3D model of the face

Since 3DMM uses the principal component analysis (PCA) method to build a statistical model, and PCA is essentially a low-pass filter, the effect of this type of method is still not ideal in restoring the detailed features of the face.

However, traditional methods tend to be slower in processing speed, or need to input additional information

Images