Face Image Lighting Enhancement Using a 3D Model

Qiulin Chen, Jan P. Allebach; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 2775-2784

Abstract


Image enhancement helps to generate balanced lighting distributions over faces. Our goal is to get an illuminance-balanced enhanced face image from a single view. Traditionally, image enhancement methods ignore the 3D geometry of the face or require a complicated multi-view geometry. Other methods cause color tone shifting or over saturation. Inspired by the new research achievements in face alignment and face 3D modeling, we propose an improved face image enhancement method by leveraging 3D face models. Given a face image as input, our method will first estimate its lighting distribution. Then we build an optimization process to refine the distribution. Finally, we generate an illuminance-balanced face image from a single view. Experiments on the FiveK dataset demonstrate that our method performs well and compares favorably with other methods.

Related Material


[pdf] [arXiv]
[bibtex]
@InProceedings{Chen_2023_CVPR, author = {Chen, Qiulin and Allebach, Jan P.}, title = {Face Image Lighting Enhancement Using a 3D Model}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2023}, pages = {2775-2784} }