Curvelet-Driven Food Image Recognition and Nutritional Analysis: A Multistage Transform Approach

Abstract

Food image recognition and nutritional estimation are now playing an increasingly important role in dietary monitoring, public health analytics, and digital health-care services. While most of the recent systems rely heavily on models of artificial intelligence, classical mathematical techniques offer value in situations where computational simplicity, interpretability, and independence from large datasets are needed. The aim of this paper is to present a comprehensive system for complete food recognition and nutrition analysis using the Curvelet Transform-a multistage, geometric analysis tool that is very effective in capturing representations of curved edges and textured patterns. In this system, the modules perform food image pre-processing, Curvelet-based feature extraction, deterministic rule-based classification, and nutritional mapping based on standardized food composition tables. Experiments conducted on 15 major food categories have demonstrated that Curvelets are well-suited for foods with strong textures and geometric structure, and the proposed approach achieves an average recognition accuracy of 86.5% and an average nutrient estimation error of 5.3%. These results confirm the potential of Curvelet-based approaches for interpretable and lightweight food analytical systems.