Improving the Accuracy in 3-Dimensionalization of Single Face Images Using 3DMM

Abstract

AI cannot identify individuals with faces wearing masks or sunglasses. This is because AI cannot recognize the eyes and mouth. If a single face image can be accurately converted to 3D, the number of features to be extracted can be increased, and the original shape of the face can be predicted even in the presence of obstacles. However, in the case of single face images with obstacles, the color and shadows of the obstacles affect the 3D face image, resulting in a 3D face image that is different from the original one. Although 3D Morphable Face Model (3DMM) can convert a single face image to 3D, there is still room for improvement in its accuracy. In this study, we investigate how to improve the accuracy of the color loss function used in the calculation of the 3D model of a 2D image.