Search

We investigate image recognition of multiple food items in a single photo, focusing on a buffet restaurant application, where menu changes at every meal, and only a few images per class are available. After detecting food areas, we perform hierarchical recognition. We evaluate our results, comparing to two baseline methods., 5 pages, 7 figures

Currently, food image recognition tasks are evaluated against fixed datasets. However, in real-world conditions, there are cases in which the number of samples in each class continues to increase and samples from novel classes appear. In particular, dynamic datasets in which each individual user creates samples and continues the updating process often have content that varies considerably between different users, and the number of samples per person is very limited. A single classifier common to all users cannot handle such dynamic data. Bridging the gap between the laboratory environment and the real world has not yet been accomplished on a large scale. Personalizing a classifier incrementally for each user is a promising way to do this. In this paper, we address the personalization problem, which involves adapting to the user's domain incrementally using a very limited number of samples. We propose a simple yet effective personalization framework which is a combination of the nearest class mean classifier and the 1-nearest neighbor classifier based on deep features. To conduct realistic experiments, we made use of a new dataset of daily food images collected by a food-logging application. Experimental results show that our proposed method significantly outperforms existing methods., Accepted to IEEE Transaction on Multimedia. http://ieeexplore.ieee.org/document/8316919/

Since the development of food diaries could enable people to develop healthy eating habits, food image recognition is in high demand to reduce the effort in food recording. Previous studies have worked on this challenging domain with datasets having fixed numbers of samples and classes. However, in the real-world setting, it is impossible to include all of the foods in the database because the number of classes of foods is large and increases continually. In addition to that, inter-class similarity and intra-class diversity also bring difficulties to the recognition. In this paper, we attempted to solve these problems by using deep convolutional neural network features to build a personalized classifier which incrementally learns the user's data and adapts to the user's eating habit. As a result, we achieved the state-of-the-art accuracy of food image recognition by the personalization of 300 food records per user.

Food has great influence on our health, but at the same time it delights us. Food plays complexed role in our life because of these different aspects, thus dietary preferences should clearly show one's characteristics. However, since large food-intake record dataset was not obtained, very few researches have attacked analyzing people's dietary preferences statistically. Today, recording food-intake, or foodlogging, is becoming popular. There are several services help people do foodlogging in easier process, which enables us to gain access to the large foodlog dataset.In this paper, we focus on analysis of dietary preferences based on nutrition intake using foodlog dataset. We have found that clustering people by average nutrition intake per one meal gives fair result, which proves analyzing preference by nutrition as fair approach. We have also proposed a Bag-of-Words based method, Bag-of-Foods, to represent one's dietary preference feature, and shown that it certainly represents users' preference with richer expression than nutrition itself.

Food related web services, such as recipe websites and food journaling apps, are rapidly gaining popularity. Data from service providers and that generated by users often coexist in these services. Compared to the former, the latter, due to its randomness and lack of organization, is often difficult to incorporate into common service features like recommendation making and associative searching. This work addresses this problem by investigating the effectiveness of multiple text embedding methods in generating distributed representations of dish titles that accurately capture the concepts of the dishes which these titles are referring to. We evaluated the practical potential of these "food vectors" with a retrieval task and a nutrition estimation task, and concluded that a method based on word embedding and LSTM is most suited for these purposes.

Food diaries or diet journals are thought to be effective for improving the dietary lives of users. One important challenge in this field involves assisting users in recording their daily food intake. In recent years, food image recognition has attracted a considerable amount of research interest as a new technology to help record users 'food intake. However, since there are so many types of food, and it is unrealistic to expect a system to recognize all foods. In this paper, we propose an optimal combination of image recognition and interactive search in order to record users 'intake of food. The image recognition generates a list of candidate names for a given food picture. The user chooses the closest name to the meal, which triggers an associative food search based on food contents, such as ingredients. We show the proposed system is efficient to assist users maintain food journals.

A meal is often called by different names depending on cultures or individuals, and a significantly large number of meal names exist in food recording data from the general public. To analyze the statistics of meals in food recording data, we need to normalize meal names; in other words, make representative names for food recording data. In this paper, we present our approach to obtaining representative names by using both food recording data and web search results. Firstly, we obtained snippets of documents from the web search results for a target meal name. Next, we performed a morphological analysis and recognized recipe-named entities (recipe NEs). Finally, we obtained the most frequent words under the condition that they were included in a candidate list of representative names. We created a dataset and evaluated our proposed method. As a result, 47.96% of Top 1 candidates and 77.32% of Top 5 candidates for generated representative names coincided with those of the ground truth (GT).

Food Log is a multimedia recording tool for producing food records for many individuals. In one year of operation, Food Log has produced more than one million food records for meals eaten by users. We found nearly 70,000 unique food records among these data. In analyzing them, one of the challenges is to extract meal categories from such a large number of records. In this paper, we propose a method for compressing a meal name into a shorter representation. First, we collect similar meal names using a k-nearest neighbor search. Next, we construct a word graph to model the relationship between the meal names and items in the database. We select representative words by identifying minimal paths in the word graph. Finally, we obtain a few words that represent categorical information about the original meal name. We applied the method to data in food records for both Food Log and the Rakuten Recipe database. Our results show that the method worked effectively for both datasets.