3D facial surface reconstruction using integrated orthographic models to approximate perspective projection model

This study develops a 3D facial reconstruction system in which the perspective projection model is approximated by applying a factorization method to the piecewise orthographic projection model. The proposed system comprises five modules. The first and second modules reconstruct the 3D facial surface using a factorization method based on an orthographic projection model. However, the facial video is taken based on the perspective projection model rather than an orthographic projection model. Thus, to compensate for the difference between the two models, the third module is developed to approximate the perspective projection model by dividing the 3D facial surface into small groups and then reconstructs each group in orthographic projection module. These reconstructed results are then integrated to form a complete 3D facial surface, which is almost as accurate as the reconstruction result using a perspective projection model. The fourth module implements a novel smoothing process for the 3D facial surface by interpolating additional vertices from the vectors of the existing 3D vertices. Finally, the fifth module utilizes a new solution to overcome the missing point problem, which is caused by occlusion at high pan rotation angles, commonly arising in 3D reconstruction applications. The experimental results show that the proposed system achieves a promising result within a relatively short time.