978-1-7281-7734-2/20/$31.00 ©2020 IEEE
A filter based genetic algorithm and neural network
technique for Image Classification
Purushottam Das
Department of Computer Sc. & Engg.
Graphic Era Deemed to be University
Dehradun, India
purushottamdas82@gmail.com
Shambhu Prasad Sah
Department of Computer Sc. & Engg.
Graphic Era Hill University
Bhimtal, India
somusuman@gmail.com
Dinesh C. Dobhal
Department of Computer Sc. & Engg.
Graphic Era Deemed to be University
Dehradun, India
dineshdobhal@gmail.com
Dikendra K. Verma
Department of ECE
Graphic Era Hill University
Bhimtal, India
dikendraverma@gmail.com
Ankur Singh Bist
Department of Computer Sc. & Engg.
Graphic Era Hill University
Bhimtal, India
ankur1990bist@gmail.com
Saurabh Pargaien
Department of ECE
Graphic Era Hill University
Bhimtal, India
saurabhpargaien@gmail.com
Abstract—We present a method for classification of images
using GA-NN approach. We will take standard image data set
from UCI repository or we can create image data set by
clicking images using camera and further extracting features
using mazda software. In either case we will be getting feature-
set of images, which will be reduced (optimized) by feature
subset selection using Mutual Information based filter
approach of Genetic algorithm. This optimal feature set will be
further classified using neural network (nprtool). Finally we
are comparing our classification results with the results of
existing Multi-SVM method.
Keywords—Classification, UCI, feature-set, feature subset
selection, genetic algorithm, mutual information, neural network,
Multi-SVM.
I. INTRODUCTION
In last few years researchers have been trying to find a
solution that prevents human errors in classification [1, 2, 3].
In the quality control system computer vision system can
replace human operator. The operator may lose focus after
working for so many hours and that may affect the evolution
process. In computer vision system, we are more prompt at
accuracy and speed. Here, we have various varieties of
grains, that make our classification process much more
complicated.
Our study displayed that the classification results are
good when we are having discriminating features of various
varieties. If we have much similarity among classes then
classification results will degrade. The objective of work
being presented here is to identify the most promising
features of the images out of the total features extracted using
feature extraction tool ( in this case mazda ). The most
discriminating set of features was short listed using the
statistical method and then classification performance was
taken using neural networks.
To classify image into different categories differentiating
features of the images should be available. In this project the
differentiating features are extracted using the Mazda
software. The feature of the input image is then matched
with the available features of different categories. If the
input features closely match with the available features then
the image can be placed into the matched category.
Classification process includes many steps such as given
below [4]: 1) Image Acquisition
2) Pre - processing
3) Feature Extraction
4) Feature Selection
5) Classification
II. LITERATURE SURVEYED
Classification is the process of categorizing samples in
their classes by using most similar and most discriminating
features [5, 6]. Classification used in every field nowadays
starting from pattern recognition to artificial intelligence,
image classification etc. [7, 8, 9]. Regression is also used for
obtaining better results [10]. Features with higher similarity
are categorized in same class and features which
discriminates each other are classified in different classes.
Further Classification used in categorizing image data set is
known as image classification. Image classification deals
with images. We can take some standard feature data set of
images from the UCI repository or we can create our own
data set by clicking images using camera of higher
resolution.
Classification can be of categorized in two types:
Supervised classification and unsupervised classification. In
supervised classification, we need a trained function while
in unsupervised learning no human training is needed [11,
12]. We are selecting only relevant features. Relevant
features refer to those features which most discriminates the
given data. Once we are having a feature data set.
Afterwards we will try to reduce this feature data set using
optimization algorithm. Here, we will be using Genetic
algorithm for this purpose [13, 14]. Moreover, we will be
using Filter approach i.e., Mutual information for optimizing
feature set. We will apply classification on the optimized
feature set, i.e., feature subset selected after optimizing [15,
16]. We have used most discriminating features for
classifying images. First these features are extracted from
image data set or we can directly take some standard feature
2020 IEEE International Conference on Advent Trends in Multidisciplinary Research and Innovation (ICATMRI) | 978-1-7281-7734-2/20/$31.00 ©2020 IEEE | DOI: 10.1109/ICATMRI51801.2020.9398498
Authorized licensed use limited to: Raytheon Technologies. Downloaded on April 14,2021 at 09:22:56 UTC from IEEE Xplore. Restrictions apply.