A Method for Correcting Distortion According to Angle Using Rendering when Creating 3D Face Model from 2D Image

This paper proposes a novel method to correct texture distortions in 3D face models generated from 2D facial images. Based on the U-Net architecture, the method utilizes 2D images captured from various angles to generate 3D objects, aiming to effectively correct distortions caused by rendering. The key contribution of this study lies in the use of both the original 2D image and the rendered image as inputs, which enhances the quality of texture reconstruction. The experiments were conducted using the CMU Multi-PIE dataset, evaluating the similarity of the face models. The experimental results demonstrate that the proposed method significantly improves distortion correction in 3D face modeling. This research presents a new direction for face modeling in virtual and augmented reality systems, and future studies could explore the potential for extending this technique through the use of various datasets and model architectures.