Deep Learning-Based Body Mass Index (BMI) Prediction Using Pre-trained CNN Models

Abstract

Automatically predicting Body Mass Index (BMI) from face images is an interesting and challenging problem in the field of computer vision. Extracted facial features are the most important requirement for estimating BMI readings. The experiment was conducted on a visual BMI database containing 4206 face images. This work studies how to detect and crop face images through image processing methods for BMI prediction. The face detector used in the image pre-processing stage is Multi-Task Convolutional Neural Network (MTCNN). This method was based on Convolutional Neural Network (CNN) and Keras framework. Throughout this research, deep pre-trained CNN models were implemented to go through the training task and to evaluate the performance of those models for the BMI prediction system. Also, the aim of this research is to predict BMI scores and BMI classes from detected face images using three pre-trained CNN models. This research provides some deep insights which are image pre-processing method using MTCNN algorithm performed better than traditional methods such as Haar classifier or Adaboost algorithm and also pre-trained CNN models resulting in better accuracy with less computation time.KeywordsDeep learningHaarAdaboostCNNMTCNNBMI