Single Food Image Database: A Comprehensive High Quality Image Dataset for Food Recognition in Artificial Intelligence

This paper introduces a diverse single-food image database, serving as a key resource for enhancing food ingredient recognition. Built with considerations of Taiwanese dietary habits and dietary survey data, this database is highly relevant to Taiwanese and Asian dietary culture while dietary globalization also makes it applicable to Western society. The database consists of 93 common foods with a total of 18,737 images, addressing the urgent need for high-quality visual data in food recognition analysis. The images are sourced from both web collection and manual photography, and their diversity and accuracy are maximized by incorporating various types, shapes, and imaging angles. Additionally, since the original AIFood dataset covered only 24 categories, this study conducted a detailed relabeling and refinement process, expanding the dataset to 93 categories. This enhancement provides a more granular food classification standard, significantly enhancing food recognition and classification accuracy. By offering a high-quality food image database, this work facilitates the development of automated food recognition. This dataset can be widely applied in hobby, digital health, and clinical fields, supporting AI humanity projects, and providing a more precise and culturally adaptive foundation for food recognition and subsequent nutrition analysis.