Full text
Toward Graph-Based Semi-Supervised Face Beauty Prediction Fadi Dornaika1,2, Kunwei Wang1,3, Ignacio Arganda-Carreras1,2, Anne Elorza1, Abdelmalik Moujahid1 1University of the Basque Country (UPV/EHU), Spain 2IKERBASQUE, Basque Foundation for Science, Spain 3Northwestern Polytechnic University, Xian, China Published in: Expert Systems with Applications, Volume 142, 2020, Article 112990, Elsevier. DOI: 10.1016/j.eswa.2019.112990. Available online 5 October 2019. Abstract Assessing beauty using facial images analysis is an emerging computer vision problem. To the best of our knowledge, all existing methods for automatic facial beauty scoring rely on fully supervised schemes. In this paper, we introduce the use of semi-supervised learning schemes for solving the problem of face beauty scoring when the image descriptor is holistic and the score is given by a real number. The paper has two main contributions. Firstly, we introduce the use of graph-based semi-supervised learning for face beauty scoring. The proposed method is based on texture and utilizes continuous scores in a full range. Secondly, we adapt and kernelize an existing linear Flexible Manifold Embedding scheme (that works with discrete classes) to the case of real scores propagation. The resulting model can be used for transductive and inductive settings. The proposed semisupervised schemes were evaluated on three recent public datasets for face beauty analysis: SCUTFBP, M 2 B, and SCUT-FBP5500. The obtained experimental results, as well as many comparisons with fully supervised methods, demonstrate that the nonlinear semi-supervised scheme compares favorably with many supervised schemes. The proposed semisupervised scoring framework paves the way to virtually all applications to adopt continuous scores instead of the usual discrete labels. Main Contributions •Introduces the first graph-based semi-supervised learning schemes for automatic face beauty prediction using holistic face descriptors and continuous scoring. •Proposes a nonlinear Flexible Manifold Embedding (NFME) adapted for propagating realvalued scores, allowing both transductive and inductive (unseen data) prediction. •Provides extensive testing on three public facial beauty datasets, demonstrating that graphbased semi-supervised approaches can outperform or match fully supervised methods, especially when labeled data are scarce. •Opens up face beauty analysis tasks (and other continuous regression tasks) to scenarios with limited annotation, broadening applicability and reducing dependence on large labeled datasets. 1
Impact of the Paper This work advances the field of computational aesthetics and facial analysis by enabling reliable face beauty prediction with minimal labeled data, leveraging information from both labeled and unlabeled samples via graph-based semi-supervised algorithms. Its nonlinear manifold methods (NFME) demonstrate that performance can match or exceed strong supervised baselines, making the approach attractive for deployment where annotation cost is high or datasets are small. Beyond face beauty, the methodology can be adapted to other continuous vision problems such as pain estimation, drowsiness detection, or age estimation, strengthening the role of semi-supervised learning for practical AI systems in biometrics, healthcare, and human-centered AI. Reference: Fadi Dornaika, Kunwei Wang, Ignacio Arganda-Carreras, Anne Elorza, Abdelmalik Moujahid. “Toward graph-based semi-supervised face beauty prediction.” Expert Systems with Applications, Volume 142, 2020, Article 112990. DOI: 10.1016/j.eswa.2019.112990 2
