Aesthetic Evaluation of Food Plate Images using Deep Learning

Abstract

This paper contributes to the experiments on using machine learning algorithms for image aesthetic evaluation, with particular focus on food plate image evaluation. Such an evaluation may be beneficial for professional and amateur cuisine makers, restaurant critiques, photographers, and travelers. We use a convolutional neural network at the training and recognition stages. The proof-of-concept experiments are arranged using our marked dataset of food plate images. Earth Mover's Distance algorithm is used for process validation. The analysis of results shows that the trained system predicts food image aesthetic value conforming to the collected human expert evaluations with good quality.