Projects

This is our project page. Here, we highlight current projects, both funded and unfunded. You can check out our past projects by clicking the link below.

Face Verification with Caricatures

A good caricature looks "more like a face than the face itself" -- Brennan, 1985

Caricatures exaggerate what makes someone unique, highlighting what makes them different from the average person. Humans are able to more quickly identify caricatured faces over realistic faces. We aim to combine research in machine learning and human perception in order to fundamentally change the future of face verification to inform both automated and human-based systems.

Funding

2019-2024

National Science Foundation

#1909707, $435K

Original grant awarded 08/2019. REU Supplement awarded 10/2020.

This material is based upon work supported by the National Science Foundation under Grant No. 1909707.

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Publications

Davis, Sara R., Bryson Lingenfelter, KevinMcElhinney, Shamik Sengupta, and Emily M. Hand. “CarVer: Setting the Standard for Face Verification with Caricatures.” 2023, IEEE International Joint Conference on Biometrics (IJCB). IEEE, 2023.

Thom, Nathan, Andrew DeBolt, Lyssie Brown, and Emily M. Hand. “DoppelVer: A Benchmark for Face Verification.” 2023, Springer International Symposium on Visual Computing (ISVC). Springer, 2023.

Lingenfelter, Bryson, Sara R. Davis, and Emily M. Hand. "A Quantitative Analysis of Labeling Issues in the CelebA Dataset." International Symposium on Visual Computing. Cham: Springer International Publishing, 2022.

Davis, Sara R., and Emily M. Hand. "Improving Face Recognition Using Artistic Interpretations of Prominent Features: Leveraging Caricatures in Modern Surveillance Systems." Intelligent Video Surveillance-New Perspectives. IntechOpen, 2022.

Lingenfelter, Bryson, and Emily M. Hand. "Improving evaluation of facial attribute prediction models." 2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021). IEEE, 2021.

Thom, Nathan, Hung Nguyen, and Emily M. Hand. "Consensus Subspace Clustering." 2021 IEEE 33rd International Conference on Tools with Artificial Intelligence (ICTAI). IEEE, 2021.

Datasets & Code

DoppelVer

ISVC, 2023

The DoppelVer dataset is a benchmark dataset designed for evaluating and advancing face verification and recognition algorithms. This dataset contains a diverse collection of facial images. The identities depicted were selected in pairs of doppelganger identities. A doppelganger is an identity who is highly visually similar to the source identity, such that they might be mistaken for one another.

Dataset: You can request the dataset by completing and submitting the End User License Agreement.

CarVer

IJCB, 2023

The CarVer dataset is a benchmark dataset designed to set the standard for face verification using caricatures and veridical images. This dataset contains a diverse collection identities with caricature and real images for each identity.

Dataset: You can request the dataset by completing and submitting the End User License Agreement.

Face Verification with Caricatures

A good caricature looks "more like a face than the face itself" -- Brennan, 1985
Caricatures exaggerate what makes someone unique, highlighting what makes them different from the average person. Humans are able to more quickly identify caricatured faces over realistic faces. We aim to combine research in machine learning and human perception in order to fundamentally change the future of face verification to inform both automated and human-based systems.

This material is based upon work supported by the National Science Foundation under Grant No. 1909707.
Title: Towards Human-Level Face Verification Performance using Distinctive Features
PI: Dr. Emily Hand
Project Period: August 2019 - July 2022

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Gender Bias in Political Writing

In this project we seek to analyze gender bias in political news articles. It is well known that document and word embeddings exhibit severe gender biases. We aim to hone in on this line of research to identify what exactly is being saved in these embeddings that makes them biased against women? Looking at political news articles, we remove all gendered terms and build gender classifiers based on document embeddings. In the image we have four different embedding visualizations. The top row show the embeddings for Breitbart news from the uncleaned data on the left and the cleaned data on the right. The bottom row similarly shows the uncleaned and cleaned embeddings from the New York Times. The clear separation before and after cleaning demonstrates that there is a fundamental difference in the way women and men are portrayed in political writing across the political spectrum.