The act of uploading facial pictures to the internet has become more common since the use of technology and of social media platforms in particular has grown.

Nowadays social media platforms offer many different tools to automatically modify and embellish images with the aim of masking or enhancing certain facial traits. Filtered images have been proven to be among the most heavily engaged photos on those sites, augmenting the popularity of these techniques.

Filters are not by definition used with the aim of compromising biometric systems but might as well affect the capabilities of those models on successfully classifying individuals since a filtered image can be taken as original.

The original images and videos present in the FFMF database were selected from the three publicly available datasets;

• 1000 gender-balanced images from CALFW, public benchmark for face verification that contains face pairs with age gaps;
  
• 1026 gender-balanced images from an uncropped version of the VIP attribute, composed of facial images annotated by gender, height, weight, and Body Mass Index (BMI);
  
• 52 videos from the COHFACE dataset, facial videos annotated with physiological signals of the subject such as their HR.

The upload of content and application of filters on social media platforms has some requirements: filters have to be applied online, manually and to one image or video at a time, there is a maximum number of seconds (s) and a restriction of 30 frames per second (fps) for an uploaded video.

Following all these prerequisites and in order to ease the process of manually applying the filters, we created 15s videos with the original images and pass those through the different social networks to be filtered.

When content is uploaded/downloaded to/from social media, it goes through operations such as image compression, resizing, and cropping which has been proved to have a negative impact on facial processing tasks. Therefore, the original images were uploaded to and downloaded from the different social media platforms obtaining compressed multimedia that can be used as a baseline in the experiments.

Our focus is set on filters that modify or distort the shape of facial biometric traits (by for example enlarging the lips or smoothening the skin) and whose effects are not easily noticeable for the naked eye. Some example filters from Instagram are displayed below:

As for Januray 2023, the filters used in the FFMF dataset creation were Thinner face, Relax! You Pretty! and Glam Grain from Instagram, Fresh vibes, Fresh light and Mellow glow from Snapchat and Belle and Spring glow from TikTok.

Any publication reproducing partially or totally this database must cite at least one of the following paper:

Mirabet-Herranz, N., Galdi, C., & Dugelay, J. L. (2022, September). Impact of Digital Face Beautification in Biometrics. In 2022 10th European Workshop on Visual Information Processing (EUVIP) (pp. 1-6). IEEE.
@inproceedings{mirabet2022impact,
title={Impact of Digital Face Beautification in Biometrics},
author={Mirabet-Herranz, Nelida and Galdi, Chiara and Dugelay, Jean-Luc},
booktitle={2022 10th European Workshop on Visual Information Processing (EUVIP)},
year={2022}

Mirabet-Herranz, N., Galdi, C., & Dugelay, J. L. (2024, August). Facial Biometrics in the Social Media Era: An in-Depth Analysis of the Challenge Posed by Beautification Filters. IEEE Transactions on Biometrics, Behavior, and Identity Science.
@article{mirabet2024facial,
title={Facial Biometrics in the Social Media Era: An in-Depth Analysis of the Challenge Posed by Beautification Filters},
author={Mirabet-Herranz, Nelida and Galdi, Chiara and Dugelay, Jean-Luc},
journal={IEEE Transactions on Biometrics, Behavior, and Identity Science},
year={2024} } }