Image Recognition with Keras: Case Studies and Applications

This assignment provides a comprehensive analysis of the Keras framework's application in image recognition. It begins with an introduction to the field, highlighting the increasing mainstream adoption of image recognition powered by deep learning, particularly convolutional neural networks, and the role of Keras in facilitating these networks. A literature review follows, summarizing key research on face detection and recognition, including works that employ Keras for tasks such as face detection, image orientation detection, and computational cost reduction. The core of the assignment comprises two case studies. The first examines a big data deep learning framework using Keras for pneumonia prediction from chest X-rays, detailing the system's preprocessing, deep convolutional neural network (DCNN) architecture, and performance analysis. The second case study focuses on a neural network system for license plate recognition, illustrating the use of Keras's DCNN capabilities. The assignment concludes with a discussion on the current state of the literature, the practical solutions, and the future potential of Keras in image recognition, emphasizing the framework's role in the field. It highlights the increasing interest in AI and machine learning.

THE USAGE OF KERAS 1
The Usage Of Keras: A Case Study On Image Recognition
My Name
Course Title
Professor name
Date

THE USAGE OF KERAS 2
The Usage Of Keras ; A Case Study On Image Recognition In Various Fields
Table Of Content
Introduction 2
Literature review 3
Case study 1 : Big Data Deep Learning Framework Using Keras : A case study of
Pneumonia Prediction 6
Case study 2: Neural Network System For Licence Plate Recognition 8
Discussion 9
Conclusion 10
References 11

THE USAGE OF KERAS 3
Introduction
The subject of image recognition has been mainstreamed in recent years and us currently being
used by not only companies but also individual consumers worldwide (Wang, Liang, & Shan,
2019). Under the wood, the image recognition artificial intelligence field is actually powered by
deep learning algorithms especially those derived from the layers of the convolution neural
networks which is an exact emulation of the human brain as it uses the same concept the visual
cortex part of the brain which is architectured to analyze images. This breakthrough in
technology has led to some innovative applications in various fields such as commerce,
automotives,manufacturing and evcen in the education sector, this has led to more sophisticated
applications of image recognition which traverse entertainment, educational and even for mobile
users (Giridhar, and, & Guruviah, 2019). Keras framework is one such algorithm that is used to
power the convolution neural network as it has a battery of deep learning layers used to analyse
the image data and give the results with a lot of certainty (Greengard, 2017). In this paper, we
analyses the literature that has been written on the subject matter using case study from two well
published works. The paper further discusses the general state of the literature and the future of
this promising technology.
Section one of this paper reviews the various literature surrounding keras. The section two uses
practical case studies to to dig deep into the uses of keras. Section three discuss the findings on
the state of the work of this framework and provided a conclusion to summarise the work .
Literature review
Most face recognition system generally is made up of the following steps; The face detection ,
extraction of feature and final the face recognition (Abiodun et al., 2018). Face detection and

THE USAGE OF KERAS 4
recognition is related and normally has complementary parts which completes the functions of
each layer. But in most instances each of the parts work independently of each other .
Over the recent years, face recognition has received widespread attention among scientists and
researchers as the field of artificial neural networks start to gain feet on the ground.(Zhao et al.,
2018) as it has been applied in widespread use cases. This section outlines some key works of
literature that has been done with respect to the various algorithms utilised.
(Cho, Ghosh, & Dehuri, 2011) presented a face detection system that uses the network of retinal
neural networks to process images by examining the image data set and make a distinction
whether the image contains a face or not. The system uses arbitrary number of networks to
improve the general performance of the algorithm. The research used three sets of training data
sets containing images. The first training set contained CMU image types, the second set
contained the scanned photography and finally the third set contained newspaper images. The
working of the algorithm utilised the Keras python framework and used the R programming
algorithms to make the classifications. The result showed a positive match identification for the
training data set with positive correlations with the desired results.
(Oyewole & Olugbara, 2018) presented a neural based image detection system. Contrary to the
classical systems which can only detect images positioned at certain angles such as upright their
framework utilised Keras to work around this limitation and feed various types of images
regardless of the orientation. The system they used deployed a set of neural networks which was
used to process the input and detect the orientation of the image and uses the output of this
preprocessing phase to choose which network to use. The overall performance of this setup from
the statistical analysis is incredible as positive matches was possible for a wide range of images
of different orientations.

THE USAGE OF KERAS 5
Most of the image recognition algorithms has a huge computational time when detecting the
human faces and has been a big hurdle in handling the facial recognition applications. In their
paper, (He et al., 2018) proposed a more fast neural network (FNN) approach that was aimed at
reducing this huge computational costs. The framework was set up by the division of the small
images and send each separate images into its layers in the convolution neural network system
which greatly utilised the Keras framework to make distinctions based on the small image which
is fed into the system. The analysis of the results showed a very high benchmarks in terms of
speed in analysing the facial data which was achieved by the FNN.
Apart from the FNN image recognition system, (Rościszewski, 2017) presented a more robust
polynomial neural network that is capable of detecting faces and images using polynomial
functions of the Keras framework. The PNN classify them into two sets of classes to make the
preprocessing decision whether it is fed with an image or non image. The framework presented
took the binomial projections of the fed images to detect faces of human beings using the
principal component analysis. The research showed the PNN when coupled with deep learning
algorithms can closely close the gap between the theoretical models and literature and yielding of
good results that can be applied in varying number of applications. In addition to the PNNs, (Cao
et al., 2018) used ANN to get the basic decision whether a preprocessed image regions has a
human face or not. The thesis described various internal working of Keras in the application
domain of the face detection applications. The thesis proposed a more hybrid approach in
handling the Keras approach algorithm which was aimed at the reduction of the number of lost
neurons in the images classified. It was however noted that the speed of the classification did not
improve significantly with performance boost of 30% recorded.

THE USAGE OF KERAS 6
All the above literature points a more active topic in the field of machine learning and neural
networks. The various frameworks such as Keras has been a breakthrough for most researchers
as they have at their disposal a very complex neural network algorithm that was primarily
designed for image classifications. So far the literature has explored the various theoretical
frameworks proposed by various scholars. The subsequent questions go much deeper by taking
two case studies and analysing the internal of the Keras framework and general image
recognition networks.
Case study 1 : Big Data Deep Learning Framework Using Keras : A case study of
Pneumonia Prediction
Nearly one million adults are diagnosed with pneumonia with surprising a fifth of the population
died from various countries globally (Bote-Curiel, Muñoz-Romero, Guerrero-Curieses, & Rojo-
Álvarez, 2019). The ability to effectively predict pneumonia is key in the medical field. The
classical method is taking the X-ray of the chest which is used to aid the doctors get the idea of
the disease hence the best prescription. The use of X-rays is a very complex task and active
research has been ongoing for better alternatives.
In their framework, (Cao et al., 2018) set up a system that utilises deep learning and neural
networks to predict whether a person is diagnosed with pneumonia or not without the use of X
rays imaging of the chest. The abstract view of this framework is as shown below.

THE USAGE OF KERAS 7
The image first preprocessed to get the standard form to be used for feeding into the neural
network and also make a check on the corrupt images which are generally discarded from the
network. The preprocessed data is then fed into the DCNN where feature extraction is done at
each level. At the end of the pipeline, a fully connected layer is output with an expected output of
1 or 0 for pneumonia or normal. The system utilised back propagation to learn from the data fed
and get the right weights. The model once trained can be used effectively to get results for
images it has never seen before (Moen et al., 2018).
The proposed solution was benchmarked for performance analysis where different parameter
was used to evaluate it. The general theme from the results showed a trusted framework that can
potentially improve the field of medicine (Wilson, 2019).

THE USAGE OF KERAS 8
Case study 2: Neural Network System For Licence Plate Recognition
Systems used for licence plate recognition has been generally receiving attention from
researchers and scientists, however, it is noted that the sensitivity of this subject is key as the
environment it operates on touches on security and privacy with complicated backgrounds of the
images for better viewing angles being a serious hurdle in having licence plate recognition
systems (Cowdrey & Malekian, 2018). The system proposed by (Yee Yong Pang, Chee Hau
Ong, & Hiew Moi Sim, 2019) uses the principle of Deep Convolutional Neural Networks(CNN)
using the R framework keras to classify images of number plates. In their model, the framework
was used with multiple layers of the DCNN to make a decision of the judgement of the number
plate images fed into the network. The DCC was programmed in the model to take feature
extraction and classifications at the same time contrary to the classical model hence do and end
to end learning. This is made possible by the multi layer approach of the Keras framework being
proposed. The model was analysed for performances was very positive and the trained data set
positive identification reached 97 % (Yee Yong Pang et al., 2019). Below is the network
workflow
Discussion
The precious sections of literature review and the two case study has revealed an active research
and practical solution being researched tested and applied in real life applications. The current
state of the literature shows the scientist have shown keen interest in artificial neural networks as
paper and thesis get published frequently. Once key framework utilised by the various models in
the R package Keras which is a very useful framework of the artificial neural network in the
image recognition domain. The package uses a cannulated layers of machine learning and deep

THE USAGE OF KERAS 9
learning codes to classify input images according to the various classification it is programmed
to do. Most algorithms uses the general steps of input feeding, pr-processing , processing and
output. The feeding is seen to be the set ups. In this phrase, the algorithm is fed with data sets to
be used for classifications. The preprocessing phase is used to remove corrupted images and
standardise the data set for efficient learning. The clean data is then fed into the convoluted
layers of deep artificial neural network to make classifications. The Keras is useful in this stage
as it has functions and methods designed to aid in the classifications. The result is a binary
decision from the classification matrix.
Conclusion
In this paper, we sought to have an understanding of the various use cases of the Keras R
framework for artificial neural network in the field of image recognition. This was achieved by
carefully analysis the existing work and literature reviewed from various researchers. The
analysis done from the literature suggests a growing number of researchers developing a keen
interest in the field of artificial intelligence as the number of literature increases. Subsequently,
the paper narrowing focused on practical case studies which includes the use of Keras framework
on the pneumonia detection and the number plate detection systems. The practical solution
supported the hypothesis of need to continuously improve on the existing literatures from a more
theoretical concepts to a much practical solutions which can positively impact our daily lives.

THE USAGE OF KERAS 10
With the increasing number of use cases of the machine learning algorithms such as Keras, the
future seems to be bright for researchers and scientists who rely on the python based algorithm to
come up with complicated classifies. Such can be improved by leveraging the power of
convolution in the multilayer approach used by Keras in making the correct feature extraction
and classifications. This is will have a lot of impact globally as more facial and image
recognition systems shall proliferate the market
References
Abiodun, O. I., Jantan, A., Omolara, A. E., Dada, K. V., Mohamed, N. A., & Arshad, H. (2018).
State-of-the-art in artificial neural network applications: A survey. Heliyon, 4(11),
e00938. https://doi.org/10.1016/j.heliyon.2018.e00938
Bote-Curiel, L., Muñoz-Romero, S., Gerrero-Curieses, A., & Rojo-Álvarez, J. L. (2019). Deep
Learning and Big Data in Healthcare: A Double Review for Critical Beginners. Applied
Sciences (2076-3417), 9(11), 2331.
Cao, C., Liu, F., Tan, H., Song, D., Shu, W., Li, W., … Xie, Z. (2018). Deep Learning and Its
Applications in Biomedicine. Genomics, Proteomics & Bioinformatics, 16(1), 17–32.
https://doi.org/10.1016/j.gpb.2017.07.003
Cho, S.-B., Ghosh, S., & Dehuri, S. (2011). Integration Of Swarm Intelligence And Artificial
Neural Network. Retrieved from http://165.193.178.96/login?url=http%3a%2f
%2fsearch.ebscohost.com%2flogin.aspx%3fdirect%3dtrue%26db%3dnlebk%26AN

THE USAGE OF KERAS 11
%3d426357%26site%3deds-live
Cowdrey, K. W. G., & Malekian, R. (2018). Home automation—An IoT based system to open
security gates using number plate recognition and artificial neural networks. Multimedia
Tools and Applications, (16), 20325. https://doi.org/10.1007/s11042-017-5407-1
Giridhar, L., and, A. D., & Guruviah, V. (2019). A Novel Approach to OCR using Image
Recognition based Classification for Ancient Tamil Inscriptions in Temples.
Greengard, S. (2017). It’s All About Image. Communications of the ACM, 60(9), 13–15.
https://doi.org/10.1145/3121434
He, L., Li, H., Holland, S. K., Yuan, W., Altaye, M., & Parikh, N. A. (2018). Early prediction of
cognitive deficits in very preterm infants using functional connectome data in an artificial
neural network framework. NeuroImage: Clinical, 18, 290–297.
https://doi.org/10.1016/j.nicl.2018.01.032
Moen, E., Handegard, N. O., Allken, V., Albert, O. T., Harbitz, A., & Malde, K. (2018).
Automatic interpretation of otoliths using deep learning. PLoS ONE, 13(12), 1–14.
https://doi.org/10.1371/journal.pone.0204713
Oyewole, S. a., & Olugbara, O. o. (2018). Product image classification using Eigen Colour
feature with ensemble machine learning. Egyptian Informatics Journal, 19(2), 83–100.
https://doi.org/10.1016/j.eij.2017.10.002
Rościszewski, P. (2017). From Linear Classifier to Convolutional Neural Network for Hand Pose
Recognition. Computer Science, 18(4), 341.
Wang, H., Liang, W., & Shan, G. (2019). An Efficient Method of Detection and Recognition in
Remote Sensing Image Based on multi-angle Region of Interests.
Wilson, A. (2019). Deep learning brings a new dimension to machine vision. Laser Focus

THE USAGE OF KERAS 12
World, 55(5), 43–47.
Yee Yong Pang, Chee Hau Ong, & Hiew Moi Sim. (2019). Deep learning Convolutional Neural
Network for Unconstrained License Plate Recognition. MATEC Web of Conferences,
05002. https://doi.org/10.1051/matecconf/201925505002
Zhao, K., He, T., Wu, S., Wang, S., Dai, B., Yang, Q., & Lei, Y. (2018). Application research of
image recognition technology based on CNN in image location of environmental
monitoring UAV. EURASIP Journal on Image & Video Processing, 2018(1), 1.