3D Face Reconstruction from Front and Profile Images for Low Computational Devices
Sankarshan Dasgupta, Mehdi R. Zargham · 2021 International Conference on Computational Science and Computational Intelligence (CSCI) · 2021
This Face reconstruction in three dimension (3D) is a difficult and advanced problem for computer vision, and our goal is to accomplish it in an efficient way to get 3D models closer to reality. Although many algorithms for construction of 3D model from two dimensional (2D) images are present, we propose an approach using front and one profile image auto-captured from video frame. The input images are used to build the UV-map and Mesh selection using various image processing techniques for small computing devices. Multiple OpenCV image processing methods are utilized for generation of the UV-map, but as its core element, it relies on the Haar Cascade face detection algorithm. For Mesh construction, it uses MediaPipe library, using Face Mesh 468 landmarks to identify the shape of the face in the image and compare it with our own dataset of mesh of most similar structure for faster 3D reconstruction. Though, we have achieved good results from the approach, we are still investigating other enhancement to make the model an identical replica.