Publications

@inproceedings{chaves_improving_2020-1,

title = {Improving Age Estimation in Minors and Young Adults with Occluded Faces to Fight Against Child Sexual Exploitation.},

author = {Deisy Chaves and Eduardo Fidalgo and Enrique Alegre and Francisco Jánez-Martino and Rubel Biswas},

url = {https://pdfs.semanticscholar.org/b1eb/3264582c54c648cd8329deddf99f64ddb094.pdf},

year  = {2020},

date = {2020-01-01},

booktitle = {VISIGRAPP (5: VISAPP)},

pages = {721–729},

abstract = {This study focuses on improving age estimation accuracy in Child Sexual Exploitation Materials, particularly for minors and young adults, using the Soft Stagewise Regression Network (SSR-Net) model. The challenge arises from unbalanced training data and facial occlusion (e.g., covering the eyes to hide victims' identities), which negatively impact the performance of age estimators. The proposed approach combines non-occluded and occluded facial images to create robust SSR-Net models. This strategy significantly enhances the model's performance, reducing the Mean Absolute Error (MAE) from 7.26, 6.81, and 6.5 to 4.07 on the IMBD, MORPH, and Deep EXpectation datasets, respectively.},

keywords = {Age Estimation, CSEM, Eye Occlusion, Forensic Images, SSR-Net-Model},

pubstate = {published},

tppubtype = {inproceedings}

}