Hair reconstruction method based on adaptive octree hair convolutional neural network

Abstract

The invention discloses a hair reconstruction method based on an adaptive octree hair convolutional neural network. The method comprises the following steps: 1) constructing a training set by using apublic hair database, converting a three-dimensional hair model in the database into a hair direction field taking a self-adaptive octree as a structure, and generating a corresponding hair directional diagram, a hair region diagram and a body region diagram; 2) designing a self-adaptive octree hair convolutional neural network to reconstruct a hair direction field; 3) preprocessing the input picture; and 4) carrying out post-processing on the output direction field, converting the direction field generated by the adaptive octree hair convolutional neural network into a hair model representedby hair, and completing hair reconstruction. According to the method, the compact structure of the octree is utilized, the storage space of the hair direction field is greatly compressed, and meanwhile, the memory overhead and the running time of a hair reconstruction network are reduced. According to the invention, the three-dimensional hair structure in the input picture can be effectively identified and recovered, and the contour appearance and texture trend consistent with those of the input hair style are ensured.

Description

technical field [0001] The invention relates to the technical field of three-dimensional reconstruction, in particular to a hair reconstruction method based on an adaptive octree hair convolutional neural network. Background technique [0002] Obtaining the 3D mesh representation of the human body surface has always been one of the important technologies in the fields of human-computer interaction, virtual reality, 3D games, animation, and movie special effects production. As hair is an important part of the human body, the importance of hair reconstruction technology is self-evident. In applications such as virtual reality, restoring realistic 3D hair models will greatly enhance the immersion of virtual environment applications and improve user comfort; the same is true in the fields of 3D games and animation film production, high-precision hair reconstruction technology will greatly enhance Realism of 3D characters. At the same time, hair reconstruction also has importan...

AI Technical Summary

Problems solved by technology

However, generating high-resolution volume data requires a large number of 3D convolution modules, and as the resolution increases, the computational overhead increases exponentially

[0004] Due to the slender and special structure of the hair, there are large empty areas in the hair direction field, that is, there are many elements whose direction vectors are zero vectors. These empty elements not only waste storage space, but also bring redundant calculation overhead to the hair reconstruction network.

Images