Personalization of food image analysis
Yuto Maruyama, Gamhewage C. de Silva, Toshihiko Yamasaki, Kiyoharu Aizawa · 2010
This paper presents a method to classify food images by updating the model of Bayesian network incrementally. We have been investigating a “food log” system which makes use of image analysis, and it can automatically detect food images and estimate the food balance (using a simple nutrition model). It also enables users to easily modify the results of the analysis when they contain errors. So far, the system does not make use of the corrections made by the users to improve the performance of classification. In this paper, we propose to incrementally update the classifier based on Baysian network so that the results of analysis will be improved by using the user's corrections. With the incremental updating, the accuracy of food image detection is improved from 89% to 92%.