Identifying the Cuisine of a Plate of Food

Ingredients are the core components that make a dish what it is, besides the preparation process. Using the idea of attribute-based classification, we seek to classify plates of food to the correct cuisine by the country, using the ingredients as attributes for a plate of food. Because of the important role that ingredients have in any plate of food, this method can be generalized to any type of dishes with any type of ingredients, and can learn new dishes not seen in the training, as long as the ingredients can be specified to fit an attribute descriptor. Our dataset came from online sources and includes three cuisines, each with two dishes represented by 76 images. Even though our dataset is limited, reasonable results and a mean accuracy of 82.9% show that the method could be generalized to more categories.