基于区域色彩迁移的3D人脸重建质量提升方法

Quality improvement method of 3D face reconstruction based on regional color transfer

针对在3D人脸重建领域内,由于3D点云数据分布不均匀,在点云数据稀疏的区域会造成颜色填充,从而导致3D人脸生成效果不佳的问题。提出了一种在PRNet的基础上基于reinhard色彩迁移算法和RGB色域分割相结合的区域色彩迁移算法。使用RGB色域分割将颜色填充部分边界确定出来,然后使用reinhard色彩迁移算法进行区域的色彩迁移。从而完成在不影响整体人脸颜色的情况下迁移指定区域色彩。在Florence和AFLW2000-3D数据集中生成的人脸质量比现有人脸生成算法生成的质量有显著的提高,为在不提高3D点云数据量的基础上改善颜色填充的情况提供了新的解决方案。

In the field of 3D face reconstruction,due to the uneven distribution of 3D point cloud data,areas with sparse point cloud data will cause color filling problems,resulting in poor 3D face generation effects. A regional color migration algorithm based on PRNet based on the combination of reinhard color migration algorithm and RGB color gamut segmentation is proposed. Use RGB color gamut segmentation to determine the boundary of the color filling part,and then use the reinhard color migration algorithm to perform regional color migration. So as to complete the migration of the specified area color without affecting the overall face color. The quality of the faces generated in the Florence and AFLW2000-3D datasets is significantly higher than that of the existing face generation algorithms. It provides a new solution for improving the color filling without increasing the amount of 3D point cloud data.