Impact of simulated glasses noise on facial emotion recognition with deep learning models
Deniz Baharvand, Fatemeh Jodati, Kimia Shirini, Sina Samadi Gharehveran · 2026
Read Impact of simulated glasses noise on facial emotion recognition with deep learning models by Deniz Baharvand, Fatemeh Jodati, Kimia Shirini, Sina Samadi Gharehveran (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13493910.txt.
Description
Baharvand and co-authors test how well emotion-reading software copes when large glasses cover a person's eyes. Systems that guess feelings from faces lean heavily on the eye region, yet real users wear chunky frames and glare, conditions lab benchmarks rarely include. In the 2026 study, the team pasted simulated oversized glasses over faces in RAF-DB, a standard emotion dataset. They ran three leading models on the altered images and released the doctored version as a public benchmark.
Accuracy fell under the occlusion for every model, confirming the eye region carries real emotional signal. Transformer-based designs, which weigh the whole face rather than small patches, weathered the damage best, with the Swin Transformer Tiny variant most robust. The convolutional ResNet-50 suffered more by comparison. The authors suggest leaning on lower-face cues when eyes are hidden.
Catalog record
| CREATOR | Deniz Baharvand, Fatemeh Jodati, Kimia Shirini, Sina Samadi Gharehveran |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |