Image-based volume estimation for food in a bowl

Authors: JIA, WENYAN - University Of Pittsburgh; LI, BOYANG - University Of Pittsburgh; XU, QI - Huazhong University Of Science And Technology; CHEN, GUANGZONG - University Of Pittsburgh; MAO, ZHI - University Of Pittsburgh; MCCRORY, MEGAN - Boston University; BARANOWSKI, TOM - Children'S Nutrition Research Center (CNRC); BURKE, LORA - University Of Pittsburgh; LO, BENNY - Imperial College; ANDERSON, ALEX - University Of Georgia; FROST, GARY - Imperial College; SAZONOV, EDWARD - University Of Alabama; SUN, MINGUI - University Of Pittsburgh

Submitted to: Journal of Food Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/4/2024
Publication Date: 1/5/2024
Citation: Jia, W., Li, B., Xu, Q., Chen, G., Mao, Z.H., McCrory, M.A., Baranowski, T., Burke, L.E., Lo, B., Anderson, A.K., Frost, G., Sazonov, E., Sun, M. 2024. Image-based volume estimation for food in a bowl. Journal of Food Engineering. https://doi.org/10.1016/j.jfoodeng.2024.111943.
DOI: https://doi.org/10.1016/j.jfoodeng.2024.111943

Interpretive Summary: Image-assisted dietary assessment has become a popular method of dietary intake in recent research. However, estimating the volume of food is still a challenging problem due to the lack of three-dimensional information in a two-dimensional image, and the blocking of the food in the image by other food or by the container (e.g., bowl, cup). This study investigated the relationship between the observable surface of food in a bowl and an index of bowl fullness to represent its volume using a mathematical model for describing different shapes of bowls. Experiments with real food/liquid demonstrated the feasibility and accuracy of the proposed approach. Such findings will help guide future research studies using this measurement concept.

Technical Abstract: Image-assisted dietary assessment has become popular in dietary monitoring studies in recent years. However, food volume estimation is still a challenging problem due to the lack of 3D information in a 2D image and the occlusion of the food by itself or container (e.g., bowl, cup). This study aims to investigate the relationship between the observable surface of food in a bowl and a normalized index (i.e., bowl fullness) to represent its volume. A mathematical model is established for describing different shapes of bowls, and a convenient experimental method is proposed to determine the bowl shape. An image feature called Food Area Ratio (FAR) is used to estimate the volume of food in a bowl based on the relationship between bowl fullness and the FAR calculated from the image. Both simulations and experiments with real food/liquid demonstrate the feasibility and accuracy of the proposed approach.