Image-based Thai Food Recognition and Calorie Estimation using Machine Learning Techniques
Rattikorn Sombutkaew, Orachat Chitsobhuk
Abstract
A wide range of health-related innovation have been developed to serve as an alternative personal health monitoring and tracking on exercise and dietary planning. The goal is to enable individuals to keep themselves healthy and disease-free. Assessing food calories helps to assist consumers in determining their calories intake each meal, leading to a strategy that regulates the amount of food they consume, and contributing to improve a control on nutrition consumption and weight loss. In this paper, we proposed a calorie estimation system on an android mobile application. Calorie estimation is performed using a food image captured from a mobile camera and the depth image from AR core library. The food area is segmented using our finetuned Mask R-CNN with Thai food image dataset. Finally, machine learning methods including Linear Regression, Support Vector Regression, K-Nearest Neighbor, and Deep Neural Network are used to estimate the amount of food calories included in a meal of each image. As a result, Deep Neural Network offers best prediction results with the most accurate prediction, the lowest error rate and the highest R-Square score.