[Paper Review] OpenFilter: A Framework to Democratize Research Access to Social Media AR Filters

Piera Riccio, Bill Psomas+4 more|arXiv (Cornell University)|Jul 19, 2022

Face recognition and analysis4 citations

TL;DR

This paper introduces OpenFilter, a framework that enables researchers to apply social media AR filters—particularly beauty filters—onto public face datasets like FairFace and LFW. By simulating real-time filter application via Android emulation, the authors create FairBeauty and B-LFW datasets, demonstrating that while beauty filters homogenize facial appearances, they preserve identity well enough to maintain performance in state-of-the-art face recognition models.

ABSTRACT

Augmented Reality or AR filters on selfies have become very popular on social media platforms for a variety of applications, including marketing, entertainment and aesthetics. Given the wide adoption of AR face filters and the importance of faces in our social structures and relations, there is increased interest by the scientific community to analyze the impact of such filters from a psychological, artistic and sociological perspective. However, there are few quantitative analyses in this area mainly due to a lack of publicly available datasets of facial images with applied AR filters. The proprietary, close nature of most social media platforms does not allow users, scientists and practitioners to access the code and the details of the available AR face filters. Scraping faces from these platforms to collect data is ethically unacceptable and should, therefore, be avoided in research. In this paper, we present OpenFilter, a flexible framework to apply AR filters available in social media platforms on existing large collections of human faces. Moreover, we share FairBeauty and B-LFW, two beautified versions of the publicly available FairFace and LFW datasets and we outline insights derived from the analysis of these beautified datasets.

Motivation & Objective

• To address the lack of publicly available datasets with AR-filtered facial images for quantitative research.
• To provide a scalable, ethical alternative to data scraping from social media platforms.
• To democratize access to AR filters for researchers across disciplines, especially in psychology, sociology, and computer vision.
• To investigate the technical and sociological effects of beauty filters on facial identity and recognition performance.
• To create and release FairBeauty and B-LFW—beautified versions of FairFace and LFW—enabling reproducible research on filter impacts.

Proposed method

• Using an Android emulator to run social media apps (e.g., Instagram) and apply real AR filters to static images from public datasets.
• Projecting dataset images onto the app’s camera feed to trigger real-time AR filter application.
• Automating the process to apply filters across large face datasets while preserving image metadata and resolution.
• Creating FairBeauty by applying beauty filters to the FairFace dataset, and B-LFW by applying them to LFW.
• Using state-of-the-art face recognition models to evaluate performance on filtered vs. original datasets.
• Simulating real social media conditions by applying multiple filters to different images, reflecting coexistence of filters in practice.

Experimental results

Research questions

• RQ1 How do beauty filters affect facial identity preservation in state-of-the-art face recognition models?
• RQ2 To what extent do AR filters homogenize facial appearances across diverse demographic groups?
• RQ3 How does the application of multiple, coexisting filters in a dataset reflect real-world social media usage patterns?
• RQ4 What are the technical and sociological implications of filter-induced aesthetic standardization?
• RQ5 How do resolution and image quality affect the visibility and impact of AR filter effects like skin smoothing?

Key findings

• Beauty filters significantly homogenize facial appearances across diverse demographic groups, reducing visual variation in the FairBeauty dataset.
• Despite homogenization, state-of-the-art face recognition models maintain high performance on the B-LFW dataset, indicating that identity is largely preserved under filter application.
• The performance of face recognition models on B-LFW shows minimal degradation, suggesting that beauty filters enhance appearance without disrupting identity recognition.
• The framework OpenFilter enables reproducible, ethical application of real social media AR filters to public datasets, overcoming limitations of data scraping.
• The resolution limit of 512x512 pixels in OpenFilter may reduce the visibility of fine filter effects like skin smoothing, potentially affecting ecological validity.
• Future research should explore how filter homogenization varies with original image similarity, particularly across age, gender, and race.