Three-dimensional image reconstruction method

Abstract

The invention provides a three-dimensional image reconstruction method, comprising steps of extracting a two-dimensional image feature point of an input image; estimating the two-dimensional image feature point by using a first mapping relationship to obtain a three-dimensional image feature point corresponding to the two-dimensional image feature point, wherein the first mapping relationship is obtained by learning a mapping relationship between an original two-dimensional image feature point and an original three-dimensional image feature point in a training sample; expanding the training sample based on a second mapping relationship, wherein the second mapping relationship is obtained by learning a mapping relationship between the two-dimensional image feature point and the three-dimensional image feature point; and acquiring a full-pixel image based on the extended training sample and the three-dimensional image feature point. The three-dimensional image reconstruction method provided by the invention can expand the training sample, and effectively solves the problem of insufficient training samples in the full-pixel image reconstruction process.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a three-dimensional image reconstruction method. Background technique [0002] In recent years, with the development of computer vision, 3D faces have been widely used, for example, in the fields of face animation, face recognition, face tracking and games. Existing 3D face reconstruction techniques can be divided into two categories: reconstruction methods based on 3D point clouds and reconstruction methods based on 2D images. Among them, the method based on 3D point cloud relies on expensive 3D data acquisition equipment and requires a high degree of cooperation from the person being collected. Therefore, this type of method is generally used in the establishment of a 3D face training library. However, methods based on 2D images can usually reconstruct 3D faces from 2D images collected daily, which expands its application range and has received extensive ...

AI Technical Summary

Problems solved by technology

[0004] The existing regression-based methods can achieve better reconstruction results, but there are also insurmountable shortcomings, such as the need for a large number of training samples to accurately learn the mapping and distribution between high-dimensional faces, but the existing 3D face training set The scale is usually small, and the types of face samples are single

Images