Calculating Body Mass Index From 2D Image Using Convolutional Neural Networks and Anthropometric Measurements

Abstract

Human body images encapsulate valuable biometric information. encompassing factors such as pupil color, gender, and weight. Of particular significance, body weight serves as a robust indicator of overall health. In alignment with recent health. science studies, this paper explores the viability of analyzing body weight through two- dimensional (2D) frontal-view human body images, utilizing the widely accepted Body Mass Index (BMI) as measure. To address varying levels of difficulty, three feasibility prob- lems are investigated, ranging from easy to hard. A comprehen sive framework is developed for body weight analysis, proposing the computation of five anthropometric features for precise body weight characterization. The correlation between the extracted anthropometric features and BMI values is thoroughly analyzed, affirming the usability of the selected features. For this study, a visual-body-to-BMI dataset is meticulously collected and cleaned, comprising 5900 images of 2950 subjects with corresponding labels for gender, height, and weight. findings demonstrate the feasibility of analyzing body The weight from 20 body images, with the proposed method outperforming two state-of-the-art facial image-based weight analysis approaches in most cases. Index Terms : Body weight analysis, visual analysis of body mass index (BMI), anthropometric features, visual-body-to- BMI dataset