DOI: 10.17798/bitlisfen.1848695 ISSN: 2147-3129

SELFIE2BFP+: An Integrated Facial Image-Based System For Body Composition, Metabolic Rate And Calorie Estimation

Yusuf Uz, Zafer Serin, Uğur Yüzgeç
Show PDF Cite
This study presents Selfie2BFP+, an integrated facial image-based assessment framework designed to estimate key body composition indicators and metabolic parameters using deep learning and mathematical modeling. Unlike conventional body composition measurement techniques that require specialized equipment or manual input, Selfie2BFP+ generates a comprehensive health profile from a single selfie image. The system incorporates three deep learning modules for gender classification, age estimation, and BMI prediction based on SmallerVGGNet, SE-ResNeXt50-32x4d, and ViT-H/14 architectures, respectively. The outputs of these models are fused through the Deurenberg equation to compute Body Fat Percentage (BFP). Subsequently, BFP, BMI, and demographic attributes are used to calculate Basal Metabolic Rate (BMR), Total Daily Energy Expenditure (TDEE), daily calorie requirements, and energy balance parameters. A web platform named SmartFit was developed to operationalize the system and provide users with real-time predictions, metabolic insights, and daily dietary activity monitoring. The platform allows users to upload or capture facial images, retrieve past predictions, visualize long-term trends, and track calorie intake and expenditure. Performance evaluations on the Face-to-BMI dataset demonstrate promising results, with MAE values of 4.51 for age estimation and 6.10 for BMI prediction, while also revealing dataset imbalance as a key limitation. Qualitative analyses indicate that the framework generates reliable proxy BFP values when upstream predictions are accurate, although deviations may occur when BMI or age estimation errors propagate. Overall, Selfie2BFP+ demonstrates the potential of combining facial analysis with metabolic modeling to support accessible, non-invasive, and multi-output health monitoring applications.

More from our Archive