Three-dimensional face reconstruction using multibaseline stereo images

Three-dimensional face reconstruction plays a crucial role in various computer vision applications, including facial expression recognition, pose estimation, virtual reality, and biometric authentication. Several methods exist for this task. However, challenges remain due to limited facial detail, training instability, difficulties handling diverse poses and expressions. This work presents a 3D face reconstruction approach using multi-baseline stereo images. First, different stereo facial images are captured by varying the baseline between cameras. Next, a dense disparity map is estimated from the stereo images with the shortest baseline. Afterward, successive disparity maps are estimated while increasing the baseline within a guided interval defined by the previous disparity estimates. Finally, the disparity map of the scene is refined and used to extract 3D information. The mathematical principles of multi-baseline stereo vision are reviewed. Moreover, experimental results for 3D object reconstruction are provided and analyzed based on objective metrics.