Face Image Lighting Enhancement Using a 3D Model

TL;DR

This paper introduces a novel face image lighting enhancement technique that utilizes 3D face models to achieve balanced illumination from a single image, overcoming limitations of previous methods.

Contribution

It leverages 3D face modeling and an optimization process to improve lighting balance in face images, addressing issues like color shifting and over saturation.

Findings

Performs well on the FiveK dataset

Outperforms existing methods in lighting balance

Effectively estimates and refines lighting distribution

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.