Method for reconstructing high-resolution human face based on grid deformation and continuous optimization

Abstract

The invention relates to a method for reconstructing a high-resolution three-dimensional human face model from some human face pictures in different angles, which comprises the following steps of: starting from two more front human face images and solving internal and external parameters of a camera and three-dimensional coordinates of an angle point by applying a motion reconstructing structure technology based on beam adjustment on the basis of the stereomatching of angle point detection (containing characteristic point); repeating the process for the human face images in adjacent angles toobtain camera parameters corresponding to all human face images; dividing the human face images based on the characteristic point and registering a human face template to the practical human face position according to the three-dimensional characteristic point; by solving the problem of nonlinear energy optimization for maintaining the consistency of human face model details and the images many times, carrying out grid deformation on the human face template to obtain the reconstructed high-resolution three-dimensional human face model; and finally, by generating a texture map, a weight map and an overall texture map corresponding to each image, carrying out texture mapping on the reconstructed three-dimensional human face model to generate a real human face.

Description

Technical field [0001] The invention relates to a method for human face modeling, in particular to a method for reconstructing a high-resolution human face model from an image based on grid deformation and continuous optimization, and belongs to the technical field of computer image and computer vision. Background technique [0002] Face reconstruction is a special problem of 3D reconstruction. It has a wide range of important applications in the fields of film and television entertainment, games, etc., and has always been valued by researchers in computer graphics, computer vision, and computer-aided design. Typical face modeling methods mainly include: manual face modeling based on CAD software, which requires a long time design process by skilled designers; and dense point cloud reconstruction based on expensive depth scanners or stereo camera equipment For three-dimensional faces, this method is costly; based on cheap video or photo-based face reconstruction, this method requ...

AI Technical Summary

Problems solved by technology

The method requires time-consuming establishment of a large number of registered face databases, and it is difficult to produce better quality face models and textures for face shapes or textures that are not in the database.

Other face reconstruction methods include methods of reconstructing faces from frontal profile images and contours, which mostly obtain lower-quality, low-resolution face models and require more cumbersome user interaction operations

[0004] There are many shortcomings in the existing face reconstruction algorithms based on images or videos, such as: many corresponding points need to be manually specified, a large number of 3D face databases are required, the reconstructed face has low resolution, and the degree of automation is low.

Images