Article Dans Une Revue Science Talks Année : 2024

Social media filters: Beautification for humans but a critical issue for AI

Résumé

Face retouching is a widespread procedure available in various modern applications. Social media provides a new tool called filters for enhancing facial images through operations like skin smoothing and virtual makeup, requiring little user expertise. These filters also deform biometric features, potentially impacting AI-based facial processing networks. Human face images encode diverse biometric information, including but not limited to gender, age, and height, in addition to traits like weight and heart rate that serve as indicators of both physical aspect and health condition. Extensive experiments revealed that filters exhibit dual aspects. On the one hand, their influence on tasks such as age, ethnicity, or weight estimation surpasses that of lossy compression, extensively studied in image processing. On the other hand, gender classifier performance improves for women with beautification filters, unraveling some form of bias, and certain filters may erase microsignals such as heart rate from videos, revealing themselves as an inexpensive facial privacy-preserving method.

Dates et versions

hal-04832069 , version 1 (11-12-2024)

Identifiants

Citer

Nélida Mirabet-Herranz. Social media filters: Beautification for humans but a critical issue for AI. Science Talks, 2024, 9, pp.100304. ⟨10.1016/j.sctalk.2024.100304⟩. ⟨hal-04832069⟩

Collections

57 Consultations
0 Téléchargements

Altmetric

Partager

  • More