Explainable Gastronomic Region Classification Using Dish Image Colors

This study researches how color in food imagery can be used to classify regional cuisines, focusing on Mesoamerican, Mediterranean, and Western North American dishes. As visual presentation becomes increasingly central to culinary marketing and tourism, understanding color patterns offers insights into regional identity and consumer preferences. We initially collected over 22,000 recipes from a popular cooking platform. Of these, approximately 14,700 included associated images and were therefore used in the image-based analysis. Regional labels were assigned using a Large Language Model, while human-interpretable color features were extracted using the Qualitative Color Description model. An interpretable Gradient Boosting classifier was trained to explore the relationship between cuisine and color. Experimental results show that color profiles vary significantly across regions, with Mediterranean dishes exhibiting the greatest chromatic complexity. Using only color-based features, the proposed model achieves a classification accuracy of approximately 76% and a macro-averaged ROC-AUC of 0.87, outperforming the stratified random baseline and consistently surpassing the alternative machine learning models evaluated in the comparative analysis. Feature importance analysis reveals that chromatic complexity and specific warm-toned hues (e.g., pale light yellow and orange) are key discriminative factors, demonstrating the model’s ability to capture meaningful chromatic signatures associated with gastronomic regions. The results confirm that an explainable, color-driven model can effectively distinguish regional cuisines, with implications for digital gastronomy, food branding, and cultural storytelling.