Food Image Recognition with Deep Convolutional Features Yoshiyuki KAWANO Keiji YANAI Abstract Department of Informatics Department of Informatics In this paper, we report the feature obtained from the The University of The University of Deep Convolutional Neural Network boosts food Electro-Communications, Electro-Communications, recognition accuracy greatly by integrating it with Tokyo Tokyo conventional hand-crafted image features, Fisher Vectors 1-5-1 Chofugaoka, Chofu-shi, 1-5-1 Chofugaoka, Chofu-shi, with HoG and Color patches. In the experiments, we have Tokyo, 182-8585 Japan Tokyo, 182-8585 Japan
[email protected] [email protected] achieved 72.26% as the top-1 accuracy and 92.00% as the top-5 accuracy for the 100-class food dataset, UEC-FOOD100, which outperforms the best classification accuracy of this dataset reported so far, 59.6%, greatly. Author Keywords food recognition, Deep Convolutional Neural Network, Fisher Vector Introduction Food image recognition is one of the promising applications of object recognition technology, since it will help estimate food calories and analyze people's eating Permission to make digital or hard copies of all or part of this work for habits for healthcare. Therefore, many works have been personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear published so far [2,4,7, 9,11]. To make food recognition this notice and the full citation on the first page. Copyrights for components more practical, increase of the number of recognizable of this work owned by others than the author(s) must be honored.