IRJET- Image Classification – Cat and Dog Images

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 570
Image Classification – Cat and Dog Images
Tushar Jajodia1, Pankaj Garg2
1Student, Department of Information Technology, Maharaja Agrasen Institute of Technology
2Assistant Professor, Department of Information Technology, Maharaja Agrasen Institute of Technology
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Many problems in computervision weresaturating
on their accuracy before a decade. However, with the rise of
deep learning techniques, the accuracy of these problems
drastically improved. One of the major problem was that of
image classification, which is defined as predicting theclassof
the image. Cat and Dog image classification is one such
example of where the images of cat and dogareclassified. This
paper aims to incorporate state-of-art technique for object
detection with the goal of achieving high accuracy. A
convolutional neural network is been build for the image
classification task.
Key Words: Image Classification, Convolutional Neural
Network, Keras, Deep Learning, Callbacks.
1. INTRODUCTION
The Dogs vs. Cats image classification has been around for a
long time now. The Dogs vs. Cats competition from Kaggle is
trying to solve the CAPTCHA challenge, which relies on the
problem of distinguishing images of dogs and cats. It is easy
for humans, but evidence suggests that cats and dogs are
particularly difficult to tell apart automatically.
Many people has worked or are working on constructing
machine learning classifiers to address this problem. A
classifier based on color features got 56.9% accuracy on the
Asirra dataset. An accuracy of 82.7% was achieved from a
SVM classifier based on a combination of color and texture
features.
In my project I am going to build a convolutional neural
network to solve the problem and achieve higher
performance and better results.
In my project instead of using the Kaggledata setcomprising
of total 25000 images, I would be working on subsetofthese
images. My dataset would be comprising of total 10000
images. Keras would used for model buildingandall thecode
would be implemented on google colab.
2. Convolutional Neural Network
The name “convolutional neural network” indicates that the
network employs a mathematical (convolution) operation.
Convolution is a specialized kind of linear operation.
Convolutional networks are simplyneural networksthat use
convolution in place of general matrix multiplication in at
least one of their layers
A convolutional neural network consists of an input and an
output layer, as well as multiple hidden layers. The hidden
layers of a CNN typically consist of a series of convolutional
layers that convolve with a multiplication or other dot
product. The activation function is commonly a RELU layer,
and is subsequently followed by additional convolutions
such as pooling layers, fully connected layers and
normalization layers, referred to as hidden layers because
their inputs and outputs are masked by the activation
function and final convolution. Thefinal convolution,inturn,
often involves backpropagation in order to more accurately
weight the end product.
Fig 1. Basic CNN Architecture
3. Dataset and Data Augmentation
The dataset in keras is divided into folders for each class.
The dataset is divided into training and testing set. The
training set and the test set compose of 2 folders one for cat
images and other for dog images. There are 4000 images of
each cat and dog for training and 1000 image of each cat and
dog for testing. The images are of varing shape and sizes, but
in order to train a CNN the images should be of same size.
Data Augmentation is being carried out by using the
ImageDataGenerator module provided by Keras.Usingitthe
images are resized to 64 x 64. Also, for training of a
convolutional neural network largesetofimagesareneeded.
So, data augmentation is appliedontheexistingsetofimages
to increase the dataset size. Various data augmentation
technique such as rescaling, shear range, zoom range are
being used to do the same.
4. Model Architecture
Classifier is the name given to the Sequential model. The
model’s first layer is a Conv2D layer. Since, it is thefirstlayer
of the model, input shape of the images that are going to be
supplied to the model is being mentioned. Next layer is a

batch normalization layer. Then one activation layer
corresponding to the conv2d layer. Further there is another
set of conv2d, batch normalization and activation layer with
different number of kernels in the conv2d layer. After that a
max Pooling layer is there and then a dropout layer is there.
The same set of layers is again repeated with different
number of kernel’s and dropout rate. The convolutionlayers
end with this set. Next are the fully connected layer.
The Sequential model API is used to build model. The
sequential API allows you to create models layer-by-layer.
The ‘add()’ function to add layers to our model. The model
needs to know what input shape it should expect. For this
reason, only the first layer in a Sequential model needs to
receive information about its input shape.
Dropout layer consists in randomly setting a fraction rate of
input units to 0 at each update during training time, which
helps prevent overfitting.
Activation layer are used to apply the activation function to
the output of that layer. The purpose of the activation
function is to introduce non-linearity into the output of a
neuron. Relu and sigmoid activation are used in the model.
Batch Normalization is used for improving the speed,
performance, and stability of artificial neural
networks. Batch normalization is a method we can use to
normalize the inputs of each layer and achieve faster
convergence.
Maximum pooling, or max pooling, is a pooling operation
that calculates the maximum, or largest, value in each patch
of each feature map. The resultsaredownsampledor pooled
feature maps that highlight the most present feature in the
patch.
In Conv2D layer, a kernel, convolution matrix, or mask is a
small matrix. It is used for blurring, sharpening,
embossing, edge detection, and more. This is accomplished
by doing a convolution between a kernel and an image. This
layer creates a convolution kernel that is convolved withthe
layer input to produce a tensor of outputs.
Then comes the fully connected layers. It contains only 2
layers. First one is the global average pooling layer to
minimize overfitting by reducing the total number of
parameters in the model. Second layer and the final layer is
the Dense layer with sigmoid activation.
Global Average Pooling 2D layer is used to minimize
overfitting by reducingthetotal numberofparametersinthe
model. GAP layers perform a more extreme type of
dimensionality reduction, where a tensor with
dimensions h×w×d is reduced in size to have
dimensions 1×1×d. GAPlayersreduceeach h×w feature map
to a single number by simply taking the average of
all hw values.
Dense layer implements the operation: output =
activation(dot( input + kernel ) + bias, activation is the
element-wise activation function passed as the
activation argument, kernel is a weights matrix created by
the layer, and bias is a bias vector created by the layer ( only
applicable if use_bias is True).
There are total 131,457 total parameters, out of which
130,881 are trainable parametersand576arenon-trainable
parameters. The major number parameters are form the
conv2d layer. Batch normalization and dense layer also
contribute few of the parameters.
5. Model Compilation and Training
During the model compilation, the optimizer algorithm, loss
function and the list of metrics are parameters which are to
be taken care of. Adam is used as the optimizationalgorithm,
binary cross entropy is used as the loss function and
accuracy is the only metric used. Early Stopping and
ModelCheckPointer callbacks areusedtopreventoverfitting
and save the best state of the model. These callbacks are
mentioned as a list during the training. Sequential models
fit_generator() is used to train the model. Model is trained
for 100 epochs with EarlyStopping and modelCheckPointer
callbacks.
6. Model Evaluation
The model trained for 15 epochs after which it stopped due
to the presence of EarlyStopping callback which had the
patience parameter set to 5. The training accuracy kept on
increasing but the validation accuracy started to decrease
which might be due to overfitting. That was the reason
EarlyStopper check pointer was used to prevent results
obtained due to overfitting. Below is the table showing the
end result of training i.e train accuracy, test accuracy, train
loss, test loss and epochs.
Table -1: Training Result Per Epoch

The graph below shows training accuracy and testing
accuracy vs the number of epochs.
Chart -1: Accuracy vs Epoch
It is evident from the graph that the train accuracy kept on
increasing with the number of epochs but the test accuracy
become near constant after the 11th epoch.
The graph below shows the train loss and test loss vs the
number of epochs.
Chart -2: Loss vs Epoch
It is evident from the graph that the train loss kept
decreasing with the number of epochs but the test loss
become near constant after 11th epoch.
The final accuracy was obtained by initializing the model
with the weights that were stored during training by the use
of ModelCheckPointer callbacks. Final accuracy of 90.10%
was obtained on the testing data.
7. CONCLUSION
In this paper we build a deep convolutional neural network
for image classification (cat and dog images). Despite of
using only a subset of the images an accuracy of 90.10%was
obtained. If the whole dataset was being used the accuracy
would have been even better.
REFERENCES
[1] Golle, P. (2008, October). Machine learning attacks
against the Asirra CAPTCHA. In Proceedings of the 15th
ACM conference on Computer and communications
security (pp. 535-542). ACM.
[2] J. Elson, J. Douceur, J. Howell and J. Saul. Asirra: a
CAPTCHA that exploits interest-aligned manual image
categorization. Proc. of ACM CCS 2007, pp. 366-374.
[3] Muthukrishnan Ramprasath, M.Vijay Anand and
Shanmugasundaram Hariharan, Image Classification
using Convolutional Neural Networks. International
Journal of Pure and Applied Mathematics, Volume 119
No. 17 2018, 1307-1319
[4] Parkhi, O. M., Vedaldi, A., Zisserman, A., & Jawahar, C. V.
(2012, June). Cats and dogs. In Computer Vision and
Pattern Recognition (CVPR), 2012 IEEE Conference on
(pp. 3498-3505). IEEE.
[5] Zeiler, M. D., & Fergus, R. (2013). Visualizing and
Understanding Convolutional Neural Networks. arXiv
preprint arXiv:1311.2901.
[6] Bang Liu, Yan Liu, Kai Zhou “Image Classification for
Dogs and Cats”.

IRJET- Image Classification – Cat and Dog Images

More Related Content

What's hot (19)

Similar to IRJET- Image Classification – Cat and Dog Images (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Image Classification – Cat and Dog Images