Comparison of Food Calorie Measurement Using Image Processing and Machine Learning Techniques

Abstract

This paper investigates the accuracy of the different Machine Learning algorithms used for calorie estimation through food image processing. Moreover, the most popular food image datasets and calorie calculation techniques will be compared. The conducted comparative study considers the accuracy of ML algorithms and the features of training and testing food image datasets. It utilizes the research efforts in the last five years to propose the most reliable combination of techniques and data to develop an efficient food image processing system. In terms of Machine Learning algorithms, the comparative study results showed that convolutional neural network (CNN) and support vector machine (SVM) algorithms are the most reliable machine learning algorithms for food image processing. Moreover, the most suitable food image datasets for training and testing food image processing ML algorithms are Food-101, UEC-Food100, and ECUSTFD. And in terms of the calorie estimation techniques, the mathematical model technique was the most efficient approach to calculating the food calorie content. The implication of this study is to guide developers to improve the current food image processing applications by using reliable algorithms, datasets, and computation strategies.