Está en la página 1de 6

BIRDS CLASSIFICATION BASED ON COLOR MOMENTS

Suryakumar H.R.
Department of Studies in Computer Science
University of Mysore
Mysore- 570006
gmail:hr.suryakumar@gmail.com

Abstract: The goal of this project is to implement a birds classification technique using an image
processing algorithm for efficient feature extraction. Color moment features like mean, standard
deviation and skewness features are extracted from the birds images and classify using KNN classifier
technique is used to classify the birds from the images. An overall 70.9% recognition rate was
achieved. Like many other methodologies different birds images may have similar or identical color
values. Hence, using color features analysis methods are still not effective enough to identify and
distinguish birds images.

I. INTRODUCTION photos. The present is the Robotic era. The


Birds are the part of nature. They robots carry pre-defined tasks efficiently with
contribute to the ecosystem and environment. embedded computer vision systems. Computer
Bird species identification from images is an vision systems have many applications such as
important and challenging problem with many performing object identification, object
applications in the real world. In order to recognition etc. The use of automated methods
evaluate the quality of our living environment it for bird species identification is an effective
is important to obtain reliable information about way to evaluate the quantity and diversity of the
the population of wild animals. Birds play birds. In all these systems, the images are
major role in Seed propagation, food chain, and processed to extract useful features that are
the beauty of nature etc. Thus, birds are greatly necessary for further recognition and
helpful to humans and nature. Hence the classification. Birds are typically categorized
identification of the birds is necessary. Recent by their shape or color and physical
machine learning and deep learning research characteristics. Bird species classification is a
has put a lot of focus on facial recognition, challenging problem both to humans and to
which involves detecting faces within images, computational algorithms because this problem
land marking specific features to account for is based on bird’s images. The majority of
the angle at which the picture was taken, and activities in the real world can be carried out
finally determining whose face is in the picture. automatically by the development of robotics.
This is extremely valuable research and has One such application envisaged is the
many applications including social media, classification of birds. A machine vision system
computer security, marketing, and healthcare. to identify and classify birds, a substantial work
While this research is booming, there has been dealing with the use of features set for
less focus on the classification of other classification of different types of birds will be
organisms like birds, for example. As with proposed for bird’s classification methodology
facial recognition, there is much complexity in from images using histograms, and other
the classification of bird’s due issues like the features like color moments. They are fed to
similarities in different subspecies of birds, various classifiers to classify them accordingly.
challenges with the foreground or background The various classifiers include Probabilistic
setting of images, and lighting conditions in Neural Network (PNN), Support Vector
Machine (SVM), Back Propagation Network KNN. Later, the processed images will be
(BPN), K-Nearest Neighbor (KNN) algorithm, further classified by using KNN (K-Nearest
K-Mean classifier, Naive Bayes Classifier. The Neighbor) classifier and finally get the
purpose of the K- Nearest Neighbors (KNN) classified label of bird from classification
algorithm is to use a database in which the data methods, which shows which bird belongs to
points are separated into several separate which group in the training sets and finally
classes to predict the classification of a new compute the accuracy. MATLAB will be used
sample point. A new bird’s classification as the programming tool for classification of
system will be proposed. These propose method birds using Image Processing toolbox.
classifies and recognizes bird’s images based
on obtained features values by using nearest
III. Feature Extraction
neighbors classification. This system also
serves as useful tool in a variety fields such as Features are distinctive attribute or aspect of
education. The challenge of such a something and feature is some essential
classification task is due to the variation in the characteristics of birds. In this project we are
background and illumination since most of the going to use features of color moments. The
images are gathered on birds’ natural habitat color moments of an image are a simple yet
and in the birds pose since it’s not possible to effective feature for color-based image
control rotation, scale, and angle of view while retrieval. From the probability theory, it is
acquiring images. Visual properties (e.g., observed that a probability distribution is
shape, color, marks, etc.) are important keys for uniquely characterized by its moments. Thus, if
bird recognition. Some researchers utilize these we interpret the color distribution of an image
features to automatically identify birds. They as a probability distribution, then the color
compared several image processing techniques distribution can be characterized by its
for the classification of bird’s species. moments as well. Furthermore, as most of the
Therefore, it is particularly important to classify color distribution information can be captured
the birds. With the development of image by the low-order moments, using only the first
identification technology, using modern three moments: mean, standard deviation and
technology to identify birds has become an skewness, it is found that these moments give a
effective research method. good approximation and have been proven to be
efficient and effective in representing the color
II. Proposed methodology distribution of images. Using these features, we
are going to classify the classes of birds.
In this project we proposed a methodology for
bird’s classification based KNN (K-nearest Mean:
neighbor classification methods. We take bird
images as training images to train the system Mean can be understood as the average color
first. To do training, initially, pre-processing value in the image
techniques will be adopted to convert the rgb
image to grayscale image and then convert
grayscale image to binary image. Then, the
preprocessed images are identified by
extracting the features like color moment Standard deviation:
feature and store in database. Then take one
The standard deviation is the square root of the
testing image and do the same preprocessing
variance of the distribution.
steps for the testing image, next step is to
compare features(values) obtained from testing
bird image with the database values using
collected totally 320 images. The dataset set
as different type of images with the
complex background. Each class has
Skewness different orientation images.

Skewness can be understood as a measure of the degree Class 1 Class 2


of asymmetry in the distribution.

Class 3 Class 4
IV. CLASSIFICATION
The classification is problem of identify the
category or class of the specified image belongs
too.

KNN (K-Nearest Neighbor) classification


K-Nearest neighbor algorithm is a method for Fig 1: Collected dataset samples
classifying objects based on closest training
examples in the feature space. K-nearest
neighbor is among the simplest of all machine
learning algorithms. Training process for the
algorithm only consists of storing feature
vectors and labels of the training images. In the
classification process, the unlabeled query point
is simply assigned to the label of its k nearest
neighbors. Typically, the object is classified
based on the labels of its k nearest neighbors by
majority vote. I k=1, the object I simply
classified as the class of the object nearest to it.
When there are only two classes, k must be an
odd integer. However, there can still be tries Fig 2: Extracted Feature values
when k is an odd integer when performing
RESULT AND ANALYSIS
multiclass classification. After we convert each
image to a vector of fixed-length with real Table 1 shows accuracies of the KNN
numbers, we used the most common distance classification with the different set of training
function for KNN which is Euclidean distance data and testing data.

V. EXPIRIMENTAL ANALYSIS AND


OBSERVATION

Dataset: I have collected the dataset from


standard dataset from ‘Keggle’ dataset. The
dataset contains 4 classes, we have
Table 1: Accuracy for KNN
properly and segment the image according to
the user need. To get more accurate different
Conclusion methodology for classification and feature
In this work, methodology of classification of extraction should be applied. We can develop a
bird’s classification based on KNN is proposed. android application for a better utilization.
This system tells us, which bird sample belong References
to which class in already stored dataset. We aim
to develop a quick, automatic and accurate [1]Anne L. Alter, Karen M. Wang “An
system for the bird’s classification. Image Exploration of Computer Vision Techniques for
processing techniques are used for extracting Bird Species Classification” published in
the color parameter like RBG, HSV, CMYK, December 15, 2017
YIQ that are having some significance and
effect on the classification of birds. Extracted [2] Fatemeh Alamdara,*, MohammadReza
features of birds based on the color and stored Keyvanpoura Computer Department,
in database and apply the k-nearest neighbor Alzahra University, Tehran, Iran “A New
method for classification. First, we do the pre- Color Feature Extraction Method Based on
processing stages and extract the components Quad Histogram” published in 2011 3rd
value of the model. In other words, the results International Conference on Environmental
of this project can be used as a specification of Science and Information Application
bird’s features that must be considered for Technology.
bird’s classification. No work has been done for
these features like color moments. So, this [3] Bahri abdelkhalak and hamid
methodology will perform well under F- zouaki,Faculty of Sciences, University
measure value and the accuracy level of the Chouaib Doukkali “Content-Based Bird
classification result is 70.9%. Retrieval using Shape context, Color moments
and Bag of Features”.

[4]John Atanbori, Wenting Duan, John


Future Work
Murray, Ko Appiah, Patrick Dickinson
In our project we have worked on the limited “Automatic classification of bird species using
feature which we have concentrated only on computer vision techniques” published in 2013
color moment feature. The goal of the project is IEEE International Conference on Image
still far. Next step in the future work are Processing.
different classifier will be tested based on our
dataset and result will be recorded. [5]Dr. Sanjay Kumar, Ankur Chauhan
Comprehensive between two or more ,Jaipur National University “Feature
classifiers will be more accurate.so that as been Extraction Techniques Based on Color
a challenging task. Due to the number of factors Images”
like image condition, illumination and image
background, most of methodology and [6]Ms.V.Anusha, Ms.V.Usha Reddy,
algorithms are unable to provide a solution for Dr.T.Ramashri M.Tech, Dept.of EEE
those problems. A consideration attention is SVUCE Tirupati, India
required to develop efficient method to “Content Based Image Retrieval Using Color
overcome these problems. Here we collect the Moments and Texture” published in
standard database so in future collect the data International Journal of Engineering Research
from itself and do the segmentation process. In & Technology (IJERT) ISSN: 2278-0181 Vol.
future can remove the background of the image 3 Issue 2, February – 2014
[7]Aastha Tiwari Banasthali University [10]S.R. Kodituwakku , Department of
,Anil Kumar DRDO Goswami Mansi Statistics & Computer Science, University of
Saraswat Banasthali University, “Feature Peradeniya“Comparison of Color Features
Extraction for Object Recognition and Image for Image Retrieval” published in Indian
Classification “ published in International Journal of Computer Science and Engineering
Journal of Engineering Research & Vol. 1 No. 3 207-211
Technology (IJERT) Vol. 2 Issue 10, October
– 2013 [11]Uma D. Nadimpalli, Randy R. Price,
[8]Ruaa Adeeb Abdulmunem Al-falluji Steven G. Hall, and Pallavi Bomma ”A
University of Babylon, Babylon, Iraq” Color Comparison of Image Processing Techniques
,Shape and Texture based Fruit Recognition for Bird Recognition”
System” published in ISSN: 2278 – 1323
International Journal of Advanced Research in [12]S.Selvarajah and S.R. Kodituwakku
Computer Engineering & Technology “Analysis and Comparison of Texture
(IJARCET) Volume 5, Issue 7, July 2016 Featuress for Content Based Image Retrieval”
International Journal of Latest Treands in
[9]Divya Srivastava, Rajesh Wadhvani and computing(E-ISSN 2045-5364) 108 Volume 2,
Manasi Gyanchandani, Research Scholar, Issue 1,March 2011
Asst. Prof. CSE Department, Maulana Azad [13] Mashael S. Aldayel, Department of
National Institute of Technology (MANIT), Information Technology King Saud University
Bhopal,M.P,india “ Color Feature Extraction Saud Arabia, Riyadh” K-Nearest Neighbor
Methods for Content Based Image” published classification for glass identification problem”
in IJCEM International Journal of Computer
Engineering & Management,Vol 18 Issue 3,
May 2015 ISSN(Online):2230-7893

También podría gustarte