Year: 2021
Type: dissertation
Abstract: Image-based dietary assessment is important for health monitoring and obesity management because it can provides quantitative food intake information. In this thesis, a novel image processing method that estimates the volume of food within a circular bowl (i.e., the top rim of the bowl is a circle) is presented. In contrast to the Western culture where circular plates are most commonly used as food containers, circular bowls are the primary food containers in Asian and African culture. This thesis focuses on estimating the volume of amorphous food (i.e., food without a clear shape, such as a bowl of cereal) instead of food with usual shapes (e.g., an apple). Four geometric features of the food, namely food orientation, food area ratio, normalized curvature and normalized shape vertex, are extracted from 2D images. Based on these features, food volume is estimated using a linear or quadratic regression model. Our experiments show that, for 135 images of six different foods in a bowl of known shape, the mean absolute percentage error of our estimation was less than 20\%, evaluated using a five-fold cross-validation technique.
Author: Boyang Li
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: green
OpenAlex ID: W3187487433