A Comprehensive Survey of Image-Based Food Recognition and Volume Estimation Methods for Dietary Assessment

Area of Science:

• Computer Science
• Health Informatics
• Nutrition Science

Background:

• Dietary problems like obesity are linked to chronic diseases, often caused by poor lifestyle choices.
• Manual food logging is imprecise, time-consuming, and has low adherence.
• Mobile health (mHealth) applications offer potential for managing dietary habits.

Purpose of the Study:

• To survey and evaluate machine learning methodologies for automatic food recognition and volume estimation.
• To discuss mobile applications implementing these automatic dietary monitoring methods.
• To identify research gaps and future directions in food image analysis.

Main Methods:

• Review of visual-based methods for automatic food recognition.
• Evaluation of methodologies using popular food image databases.
• Analysis of machine learning techniques, particularly deep neural networks and convolutional neural networks (CNNs).

Main Results:

• Deep neural networks utilizing visual features are used in approximately 66.7% of surveyed studies for food recognition.
• Convolutional neural networks (CNNs) are universally employed for ingredient recognition.
• Mobile applications are increasingly integrating these advanced methods for automatic food logging.

Conclusions:

• Automatic food recognition via machine learning shows promise for improving dietary monitoring accuracy and adherence.
• Further research is needed in areas like unsupervised learning, continual learning, and explainable AI for enhanced food image analysis.
• These advancements can significantly contribute to managing lifestyle-related chronic diseases.