A Single View 3D Face Reconstruction Method Based on Spherical Coordinate Position Map

Abstract

The invention discloses a method for reconstructing a single-view three-dimensional human face based on a spherical coordinate position map, which comprises the steps of: 1) inputting a single view containing a human face, and detecting the position of the human face in the picture; 2) cutting according to the detected position The face image on the single view is converted into a fixed size and preprocessed; 3) the preprocessed image is input to the encoder, and the encoder outputs the feature map; 4) the extracted feature map is input to the decoder, The decoder outputs a position map represented by spherical coordinates; 5) Convert the position map represented by spherical coordinates into a Cartesian coordinate position map, and reconstruct a 3D face from the converted position map. The method of the invention has the advantages of fast calculation speed, low consumption of memory resources, good generalization performance of the method, and ability to accurately reconstruct human faces in complex environments.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a three-dimensional face reconstruction method based on a spherical coordinate position map. Background technique [0002] 3D face reconstruction refers to the process of extracting face information from a single or multiple 2D face images, and reconstructing a 3D face based on the face information. 3D face reconstruction has a very high application prospect in the fields of criminal investigation, entertainment games and medical images. [0003] The human face is composed of various complex biological tissues, such as bones, muscles, eyes and other organs. The human skull determines the outline of the human face, the skin reflects the appearance of the human face, and the coordination and deformation of various muscle tissues produce various expressions of the human face. And the illumination of the face is very complicated. Even for the same face, due to the influen...

AI Technical Summary

Problems solved by technology

[0012] Specifically, in order to make the 3D face reconstruction more detailed and not limited by the model, the regression model used in the present invention is trained based on the data set represented by the position map, and the primary problem to be solved for most of the failed reconstruction cases is The prediction deviation of the face deflection angle is too large, and the representation form of the vertices in the position map is not Cartesian coordinates, but the spherical coordinates are used to represent the position of the vertices in space.

Images