Feature Extraction using GLCM for Dietary Assessment Application
Akshada A. Gade, Arati J. Vyavahare
P.E.S’s Modern College of Engineering Pune, Maharashtra, India
Abstract
This paper offers technique for dietary assessment
towards mechanically detect the type of food from
various pictures captured during eating occasions.
Recognition of food is complicated procedure since
most of the food items are varies in shape and
appearance. To achieve this task segmentation is
important for labelling of food. The features of each
segmented regions are extracted by capturing visual
content of image. System works well on the most
relevant six statistical parameters or texture features
computed by using Gray Level Co-occurrence Matrix
(GLCM). Then construct a feature vector to
represents all feature values. The operation of
classification will be performed on the basis of
defined features. Experimental results on various food
items are obtained. This food recognition system can
be easily integrated into dietary assessment
applications. By analysing food portion and size
information, system will also calculate calories and
nutrition values. For obtaining better performance
and accuracy in food recognition, system needs to
extract multiple features.
1. Introduction
Dietary intake offers significant insights for
organizing interference programs for prevention of
sickness and disease. As there is a rising worry with
regard to unending diseases or several health
problems interrelated to the diet together with obesity
or cancer so one needs to do correct judgment of
foodstuff and beverage consumption. It is essential to
promote individuals to be healthy and maintain great
lifestyles by participating in the control of long-time
period fitness selections. Determine accurate dietetic
intake is considered to be a difficult challenge in the
health and nutrition fields [8]. Traditional dietary
assessment is comprised of written and orally mention
strategies which is tedious and time consuming, also
requires nutrition expert for everyday monitoring.
This project builds an approach to locate and
recognize perceptually similar food objects with the
help of single food image where every food item is
identified with the help of segmentation and image
features for dietary assessment applications. The
research also focuses on providing correct number of
calories and nutrient intake. This method makes use
of the image analysis tools for dietary assessment
application.
2. Literature survey
Dietary assessment has been a trendy topic in
biomedical and health associated fields for years. In
case of computer vision, food identification is a
category of recognition. Fengqing Zhu [1] proposed a
Multiple Hypothesis Segmentation and Classification
(MHSC) system where food identification is
accomplished by integrating features such as texture,
color and SIFT descriptors. These features combined
to form single feature vector and classification is done
by using Support Vector Machine (SVM) which
provides 65% accuracy on Japanese food databases.
Edward J. Delp and carol J. Boushey [2] described
new approach for food identification using
combination of 8 local features and 4 global features
for accurate visual description of food item. A
“voting” based late decision fusion classifier used to
identify the food items and by using feature channel
one can improve the classification rate more than 7%.
Marios M. Anthimopoulos [3] develops a food
recognition system used to estimate meals
carbohydrate content especially for diabetic patients
which is based on bag-of-feature (BoG) model.
Features are extracted with the help of SIFT descriptor
or color descriptor. The results say that the SIFT-
based descriptors are comparatively less sensitive to
color shifts and intensity changes so their
performance is much better than the color descriptors.
In 2016, Ju-Chin Chen [4] proposed a system based
on nutrition composition analysis by using local
orientation descriptor, where the features of the food
deal with relative to variations in scale, texture,
rotation, and deformation. Local Orientation
Descriptor (LOD) and colour features such as colour
moment and histograms in YIQ colour space are
incorporated for classification. For nutrition
classification SVM with RBF kernel is applied. For
quantity estimation coin used as a reference object.
For the Japanese food database combination of color
features and LBP system gives food recognition rate
of 82.8%, combination of color and texture features
gives recognition rate up to 85.6% and color features
with LOD system gives highest food recognition rate
as 87.9%. Shijin Kumar and Dharun [5] shows the
extraction of nine texture features using GLCM and
three shape Features such as perimeter, area,
circularity using Connected Regions for diagnosis of
brain abnormalities, where they conclude that the
International Journal Multimedia and Image Processing (IJMIP), Volume 8, Issue 2, June 2018
Copyright © 2018, Infonomics Society 409