Page 133 - Petelin, Ana, et al. Eds. 2019. Zdravje otrok in mladostnikov / Health of Children and Adolescents. Zbornik povzetkov z recenzijo / Book of Abstracts. Koper: Založba Univerze na Primorskem/University of Primorska Press
P. 133
t is visible on the photography of food and beverages? ikt rešitve in storitve | ict solutions and services
Simon Mezgec1, Tome Eftimov2,3 Tamara Bucher4, Barbara Koroušić Seljak21
Jožef Stefan International Postgraduate School, Jamova cesta 39, 1000 Ljubljana,
Slovenia
2 Jožef Stefan Institute, Jamova cesta 39, 1000 Ljubljana, Slovenia
3 Stanford University, Stanford, CA 94305, USA
4 University of Newcastle, Callaghan, NSW 2308 Newcastle, Australia
Introduction: Eating habits are usually monitored by keeping a food diary, i.e.
by precisely recording the consumed food and drinks. Today, there is a number
of applications for keeping an electronic food diary, which facilitate the track-
ing of eating habits, but they require a lot of discipline in weighing and record-
ing consumed meals.
Methods: In the paper, we will present a state-of-the-art computer technolo-
gy for automated recognition of food and beverages from photographs. The
technology is based on a deep neural network for image recognition and rep-
resents an upgrade over existing automated solutions. Evaluation of eating hab-
its requires, in addition to recognizing food and drink images, the mapping of
this information to a food composition database. This can also be obtained in
an automated way by using natural language processing.
Results: The neural network was trained to recognize images using data col-
lected in a study on fake foods. The network is currently able to identify 55 dif-
ferent food classes with a high degree of accuracy (92.18 %). The degree of ac-
curacy of the mapping to the food composition database is 93 %.
Discussion and conclusions: We can integrate this technology into a mobile
app, with which a user captures photos of food or beverages. The network is
constantly learning from new user photos, which means it will be even more
accurate in the future and will be able to identify more classes of food and bev-
erages.
Keywords: food image recognition, nutritional value, fake food
131
Simon Mezgec1, Tome Eftimov2,3 Tamara Bucher4, Barbara Koroušić Seljak21
Jožef Stefan International Postgraduate School, Jamova cesta 39, 1000 Ljubljana,
Slovenia
2 Jožef Stefan Institute, Jamova cesta 39, 1000 Ljubljana, Slovenia
3 Stanford University, Stanford, CA 94305, USA
4 University of Newcastle, Callaghan, NSW 2308 Newcastle, Australia
Introduction: Eating habits are usually monitored by keeping a food diary, i.e.
by precisely recording the consumed food and drinks. Today, there is a number
of applications for keeping an electronic food diary, which facilitate the track-
ing of eating habits, but they require a lot of discipline in weighing and record-
ing consumed meals.
Methods: In the paper, we will present a state-of-the-art computer technolo-
gy for automated recognition of food and beverages from photographs. The
technology is based on a deep neural network for image recognition and rep-
resents an upgrade over existing automated solutions. Evaluation of eating hab-
its requires, in addition to recognizing food and drink images, the mapping of
this information to a food composition database. This can also be obtained in
an automated way by using natural language processing.
Results: The neural network was trained to recognize images using data col-
lected in a study on fake foods. The network is currently able to identify 55 dif-
ferent food classes with a high degree of accuracy (92.18 %). The degree of ac-
curacy of the mapping to the food composition database is 93 %.
Discussion and conclusions: We can integrate this technology into a mobile
app, with which a user captures photos of food or beverages. The network is
constantly learning from new user photos, which means it will be even more
accurate in the future and will be able to identify more classes of food and bev-
erages.
Keywords: food image recognition, nutritional value, fake food
131
