International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022
p-ISSN: 2395-0072
www.irjet.net
Review on Image-based Body Mass Index Prediction Methods Sruthy R1 1Guest. Lecturer, Dept. of
Electronics and Communication Engineering, NSS Polytechnic College, Kerala, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract –An individual's body mass index (BMI), is a
this type of model, an estimation of BMI is based on a single or pair of two-dimensional body images. This model maps weight/BMI values based on five anthropometric features taken from body images. The method can categorize the weight variance from pairs of photos. The classification task is done by Multi Support Vector Machine (SVM) and BMI prediction is done by Support Vector Regression (SVR).
measure based on height and weight, that assesses a person's level of body fat. It is a tool to identify whether an individual is underweight, overweight, or healthy. Health risks may increase if a person's BMI is not within the healthy range. Traditional BMI assessment is prone to accuracy issues. This has a negative impact on determining a person's fitness. With the emergence of technologies, several image processing techniques have been developed to estimate body mass index from human body images. This paper is a review of different image-based BMI prediction methods.
A facial image-based BMI prediction system has been implemented in [4]. Face Net and VGG face and Region aware Global Average Pooling method are used to extract the facial features from the images. Then the feature vectors obtained are fed to the regression module to predict the body mass index.
Key Words: Body Mass Index (BMI), Anthropometric features, Conditional Random Field Recurrent Neural Network (CRF-RNN), Support Vector Machine (SVM), Support Vector Regression (SVR), Region aware Global Average Pooling method (Re-GAP) Magnetic Resonance Image (MRI), Gradient-Weighted Class Activation Mapping (Grad-CAM), Rectified Linear Unit (ReLu), Residual Network(ResNet)
In [5], a deep learning-based approach has been employed to estimate body mass index from structural Magnetic Resonance images (MRI). The CNN localization maps revealed the caudate nucleus and the amygdala as brain regions that contributed strongly to BMI prediction. Gradient information from the final layer of the CNN is used in a method known as gradient-weighted class activation mapping (Grad-CAM)[6] to identify brain areas that are significantly connected to BMI prediction.
1. INTRODUCTION The ratio between the weight and height of a person can be expressed as a number called body mass index (BMI). It could provide a fundamental understanding of someone's weight status [1]. It is measured commonly using a device called scale or balance. Several use cases require an assessment of body weight without the direct presence of the person. With the advancement of image processing methods, it is now possible to predict BMI from body images without knowing a person's height or weight.
In [7], BMI is assessed from silhouette images utilizing convolutional neural networks. Silhouettes represent objects, people, or scenes in one color, typically black, with edges corresponding to the contours of the subject [8]. For estimating BMI, six machine learning models and MiniVGGNet deep learning model are employed. When these 7 models' performance is compared, it is discovered that CNN outperforms the other 6 machine learning models. The 6 machine learning models are Gabor-RF (Random Forest), HOG (Histogram of Gradient)-RF, GaborXGB(XGBoost), HOG-XBG, Gabor-SVM (Support Vector Machine), and HOG-SVM.
A person's weight can be difficult to measure using a weighing machine when they are in medical emergencies because they can't be moved. A weight estimation could have a great deal of value in forensic science as well. An automatic search through video surveillance records could utilize weight along with other physical traits to describe the fugitive [2]. This article discusses various image processing methods that can be used as an alternative approach to accurately calculate a person's body mass index. In Chapter 3, the different BMI estimation methods are discussed. The comparison of these approaches is presented in Chapter 4. The conclusion of the work is given in Chapter 5.
A deep ResNet model for determining Body Mass Index has been given the depth image of the individual obtained from the Kinect Sensor in [9]. In order to train the deep neural network model to predict BMI scores, the suggested method generates a significant amount of synthetic depth pictures of virtual manikins. Making use of the MakeHuman software, 3D manikins are developed. Body mass index has been determined from facial image features using the logistic regression model in [10]. The method makes use of two distinct datasets that are composed of different kinds of face image data. Seven essential
2. LITERATURE REVIEW A computational model for BMI estimation using twodimensional human body images has been proposed in [3]. In
© 2022, IRJET
|
Impact Factor value: 7.529
|
ISO 9001:2008 Certified Journal
|
Page 1119