Dietary assessment is paramount for individual health and plays a significant role in preventing and managing chronic diseases. Traditional methods of tracking food intake, such as recall and manual logs, are often plagued by inaccuracies, especially over extended periods. Emerging technologies utilizing mobile cameras and computer vision offer a promising alternative, streamlining the process and enhancing the precision of diet monitoring. This article delves into the advancements in Food Image-based recognition systems (IBFRS), which leverage smartphone cameras and computer vision techniques alongside publicly accessible food datasets.
These innovative systems typically operate in stages. Initially, food images captured by a mobile device undergo segmentation to isolate individual food items on a plate. Subsequently, these items are classified into specific food categories using sophisticated algorithms. Finally, the system estimates the volume, caloric content, and nutritional value of each food item, providing a comprehensive dietary analysis.
A thorough review of 159 studies in the field of IBFRS reveals significant progress. Among the 78 studies meeting stringent inclusion criteria, a detailed examination of methodologies and performance metrics on publicly available food datasets was conducted. Notably, studies lacking performance data for at least one key phase of IBFRS were excluded to maintain rigor. Deep learning methodologies, particularly Convolutional Neural Networks (CNNs), are increasingly dominant, being employed in 58% of the analyzed studies. The superior performance of CNNs on large food datasets underscores their efficacy, as these datasets provide the necessary resources for effective algorithm training.
The practical benefits of IBFRS extend to professional dietetic practice. These systems offer tools for more objective and efficient dietary data collection, potentially transforming how dietitians monitor and advise patients. However, challenges remain in the widespread adoption of IBFRS. These include improving accuracy in diverse and real-world eating scenarios, addressing variations in food image quality due to lighting and angles, and ensuring user-friendliness for consistent data capture. Future research directions are focused on overcoming these limitations, further refining algorithms, and expanding the accessibility and applicability of food image analysis in dietary assessment.
In conclusion, food image recognition systems represent a significant leap forward in dietary assessment technology. By automating and improving the accuracy of food intake monitoring, IBFRS holds the potential to revolutionize both personal dietary management and professional nutritional guidance, ultimately contributing to improved health outcomes.
