3D Face Reconstruction from 2D Images Using CNN

Munireddy M S, Suprit V Hatti, Tarun Siddappagoudar, Pavan C Karaveeramath, Channabasappa Muttal

2025

Abstract

3D face reconstruction from 2D images is a significant challenge in computer vision, with applications in augmented reality, biometrics, and healthcare. Our framework starts with robust facial landmark detection to localize key regions such as the eyes, nose, and mouth, enabling the initialization of a parametric BaseFaceModel. This model, based on a 3D morphable face model (3DMM), compactly encodes facial shape and expression. To enhance realism, the BaseFaceModel is refined using an AlbedoFaceModel, which reconstructs the albedo by disentangling lighting effects from the image. These refined models provide the foundation for our deep convolutional neural network reconstruction pipeline. The pipeline integrates a three-stage loss function: geometric loss ensures structural consistency with landmarks, photometric loss minimizes pixel-level differences, and perceptual loss captures high-level semantic details. Moreover, a skin mask generation step improves texture quality and reconstruction precision. Experimental results show a landmark detection accuracy of 94% and reconstruction accuracy of 81%. By combining these advanced modeling techniques with a tailored loss framework, this approach delivers a robust, high-fidelity workflow for 3D facial reconstruction, offering immense potential for applications requiring precise 3D face modeling.

Paper Citation

in Harvard Style

M S M., Hatti S., Siddappagoudar T., Karaveeramath P. and Muttal C. (2025). 3D Face Reconstruction from 2D Images Using CNN. In Proceedings of the 3rd International Conference on Futuristic Technology - Volume 3: INCOFT; ISBN 978-989-758-763-4, SciTePress, pages 286-294. DOI: 10.5220/0013614200004664

in Bibtex Style

@conference{incoft25,
author={Munireddy M S and Suprit V Hatti and Tarun Siddappagoudar and Pavan C Karaveeramath and Channabasappa Muttal},
title={3D Face Reconstruction from 2D Images Using CNN},
booktitle={Proceedings of the 3rd International Conference on Futuristic Technology - Volume 3: INCOFT},
year={2025},
pages={286-294},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013614200004664},
isbn={978-989-758-763-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 3rd International Conference on Futuristic Technology - Volume 3: INCOFT
TI - 3D Face Reconstruction from 2D Images Using CNN
SN - 978-989-758-763-4
AU - M S M.
AU - Hatti S.
AU - Siddappagoudar T.
AU - Karaveeramath P.
AU - Muttal C.
PY - 2025
SP - 286
EP - 294
DO - 10.5220/0013614200004664
PB - SciTePress