Assessment of AI Technologies Applied to Food Image Calorie Estimation

Abstract

Artificial intelligence methods are increasingly used to support food recognition, portion estimation and calorie assessment from digital images. This review examines selected AI technologies applied to food-image calorie estimation, with attention to food detection, segmentation, recognition, volume estimation and dataset availability. It discusses earlier mobile and cloud-based systems, machine-learning approaches such as support vector machines, and deep-learning methods, including convolutional neural networks, Faster R-CNN, semantic segmentation models and few-shot learning approaches. The review also summarises commonly used food-image datasets designed for food recognition, detection, segmentation and volume estimation. The analysis shows that AI-based calorie-estimation systems can improve convenience by reducing manual food logging and supporting automated nutritional assessment. However, accuracy depends on several linked tasks, including correct food classification, precise segmentation, reliable portion-size estimation and appropriate nutritional mapping. Food-recognition accuracy can be high under controlled conditions, but calorie estimation remains more difficult because food volume, density, preparation method, lighting, occlusion and mixed-food presentation introduce substantial uncertainty. Cloud-based processing can support computationally demanding models, while object-detection and segmentation methods can improve analysis of complex meals. Nevertheless, current systems still face important limitations in real-world food images. The review concludes that AI technologies provide a promising foundation for automated calorie estimation, but further work is required to improve dataset quality, depth estimation, contextual food information, local nutrition databases and validation under practical eating conditions.