When Age-Invariant Face Recognition Meets Face Age Synthesis: A Multi-Task Learning Framework and a New Benchmark

Abstract

To minimize the impact of age variation on face recognition, age-invariant face recognition (AIFR) extracts identity-related discriminative features by minimizing the correlation between identity-and age-related features while face age synthesis (FAS) eliminates age variation by converting the faces in different age groups to the same group. However, AIFR lacks visual results for model interpretation and FAS compromises downstream recognition due to artifacts. Therefore, we propose a unified, multi-task framework to jointly handle these two tasks, termed MTLFace, which can learn the age-invariant identity-related representation for face recognition while achieving pleasing face synthesis for model interpretation. Specifically, we propose an attention-based feature decomposition to decompose the mixed face features into two uncorrelated components-identity-and age-related features-in a spatially constrained way. Unlike the conventional one-hot encoding that achieves group-level FAS, we propose a novel identity conditional module to achieve identity-level FAS, which can improve the age smoothness of synthesized faces through a weight-sharing strategy. Benefiting from the proposed multi-task framework, we then leverage those high-quality synthesized faces from FAS to further boost AIFR via a novel selective fine-tuning strategy. Furthermore, to advance both AIFR and FAS, we collect and release a large cross-age face dataset with age and gender annotations, and a new benchmark specifically designed for tracing long-missing children. Extensive experimental results on five benchmark cross-age datasets demonstrate that MTLFace yields superior performance than state-of-the-art methods for both AIFR and FAS. We further validate MTLFace on two popular general face recognition datasets, obtaining competitive performance on face recognition in the wild. The source code and datasets are available at http://hzzone.github.io/MTLFace.

Facial Recognition Technology for Identifying Missing Individuals and Wanted Criminals

“…These comparisons have been conducted with both DC-GAN and α-GAN face generators being well-tuned. In [15],The paper suggests an identity conditional block (ICB) to achieve identity-level aging/rejuvenation pattern and a weightsharing strategy to enhance the age smoothness of synthesized faces in order to address these problems brought about by one-hot encoding. To be more precise, the suggested ICB learns an identity-level aging/rejuvenation pattern by using the identity-related feature from AFD as input.…”

HP-FLDet: A High Performance Joint Face and Landmark Detector

“…With the face detection heads on the top right side in M indicates the number of Gaussian distribution of mixture dense network in equation (6). Therefore, the overall multi-task loss is composed of 4 components including the face classification, the facial bounding box localization, the heatmap-based landmark localization and the regressionbased landmark localization as in equation (1).…”

“…Face detection and facial landmark detection are two essential and fundamental steps in many facial analysis applications, such as recognition [1,2], generation [3], manipulation [4], tracking [5], attributes analysis [6], etc. The purpose of face detection is to locate facial bounding boxes on images, while facial landmark detection seeks to predict landmarks (i.e., nose, eye, etc.)…”

GFAM: A Gender-Preserving Face Aging Model for Age Imbalance Data

“…Liu et al [21] proposed a framework that learns to separate age-related features from identity-related features in face images to generate aging faces while preserving identity. Huang et al [22] proposed a multi-task learning framework that combines age-invariant face recognition with face age synthesis and utilized these high-quality synthesized faces to further boost age-invariant face recognition. Huang et al [23] proposed a framework that integrates the advantages of a flow-based method model and GANs.…”

“…Wang et al [12] 2018 RNN Autoencoder CAAE [1] 2017 cGAN Autoencoder IPCGAN [2] 2018 cGAN Identity-Preserved AMGAN [3] 2020 cGAN High Resolution Dual conditional GAN [4] 2018 cGAN Dual Conditional GANs dual AcGAN [5] 2020 cGAN Spatial Attention Mechanism PAG-GAN [6] 2018 GAN Pyramid Architecture Li et al [15] 2018 GAN Global and Local Features WGLCA-GAN [16] 2019 GAN Wavelet Transform A3GAN [8] 2021 GAN Attention Mechanism Yao et al [17] 2020 GAN High Resolution S2GAN [19] 2019 GAN Share Aging Trends PFA-GAN [7] 2020 GAN Progressive Neural Networks Re-Aging GAN [20] 2021 GAN Face Age Transformation Liu et al [21] 2021 GAN Disentangled Representation MTLFace [22] 2022 GAN Multi-task Learning AgeFlow [23] 2021 GAN Flow-based Zhao et al [24] 2022 GAN Child Face Prediction…”