Facial reshaping operator for controllable face beautification

Shanfeng Hu, Hubert P.H. Shum*, Xiaohui Liang, Frederick W.B. Li, Nauman Aslam

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Posting attractive facial photos is part of everyday life in the social media era. Motivated by the demand, we propose a lightweight method to automatically and efficiently beautify the shapes of both portrait and non-portrait faces in photos, while allowing users to customize the beautification of individual facial features. Previous methods focus on the beautification of mostly frontal and neutral faces, without incorporating user controllability in the beautification process. To address these restrictions, we propose the Facial Reshaping Operator representation, which is affine-invariant, captures the pairwise geometric configuration of facial landmarks, and allows for efficient face beautification with the user-specified weights of individual facial parts. We also propose an unsupervised beautification method in the operator space of faces, where an input face is iteratively pulled towards a local nearby density mode with improved attractiveness. Our method distinguishes itself from the commercial beautification tools in that it mildly enhances facial shapes without altering makeups or complexions, which complements these tools that lack fine-grained control on the attractiveness of facial shapes for users. The experimental results show that our method improves facial shape attractiveness for a large range of poses and expressions, demonstrating the potential of applicability to photos seen on the social media such as Facebook and Instagram everyday.
Original languageEnglish
Article number114067
Number of pages13
JournalExpert Systems with Applications
Volume167
Early online date8 Dec 2020
DOIs
Publication statusPublished - 1 Apr 2021

Fingerprint

Dive into the research topics of 'Facial reshaping operator for controllable face beautification'. Together they form a unique fingerprint.

Cite this