Image Recognition Using Deep Learning: A Detailed Report

This report delves into the application of deep learning for image recognition, a crucial aspect of modern technology. It begins with an introduction to the field, emphasizing the challenges of automatic image classification and the role of deep learning algorithms like Convolutional Neural Networks (CNNs). The report discusses image classification using deep learning, highlighting the advantages of CNNs in feature extraction and image analysis. It then explores the use of deep learning in medical image processing, emphasizing its potential to revolutionize healthcare through automated diagnosis and analysis. Furthermore, the report examines the application of image analytics in various businesses, showcasing how deep learning can be utilized for efficient data extraction and decision-making. The report concludes with a summary of the key findings and the future prospects of deep learning in image recognition.

Running head: IMAGE RECOGNITION USING DEEP LEARNING
IMAGE RECOGNITION USING DEEP LEARNING
Name of student
Name of university
Author’s note:

1
IMAGE RECOGNITION USING DEEP LEARNING
Abstract
In the present technological world, with the introduction of several modern technologies, the
requirement of possessing the modern technologies has increased significantly. This report
intends to discuss the utilisation of the deep learning algorithm to execute image detection.
The algorithms of deep learning has helped several companies with image recognition and
gain the data from images. A brief discussion of Image classification using deep learning has
been provided. The discussion of extensive deep learning for the medical image processing
has been provided in this report and the Application of image analytics for the businesses has
been discussed briefly in the report. This report concludes with an appropriate conclusion for
the report.

2
IMAGE RECOGNITION USING DEEP LEARNING
Table of Contents
Introduction....................................................................................................................3
Discussion......................................................................................................................4
Image classification using deep learning...................................................................4
Deep learning for the medical image processing.......................................................7
Application of image analytics for the businesses.....................................................9
Conclusion....................................................................................................................10
References....................................................................................................................11

3
IMAGE RECOGNITION USING DEEP LEARNING
Introduction
In the recent years, along with the significantly speedy development in the
identification of the digital content, the automatic classification of images grew to be among
the most challenging task in sector of the computer vision. The automatic comprehending and
the analysis of the images by the system is significantly challenging as compared to the
human visions. Several research has been executed for overcoming the issues in the
prevailing classification systems, but main output has been narrowed solely for the low level
image primitives. Moreover, these approach mainly lacks with the appropriate classification
of the images. The main researches in the image processing associates with the use of the
deep learning methods for achieving the desired results in area such as the computer visions.
The system of the earlier researches mainly presents the utilisation of the Convolutional
Neural Networks, the algorithm of machine learning that is being utilised for automatic
classification of the images.
This paper discusses the modern methods of image processing using the techniques of
deep learning and the neural networks. The deep learning has been significantly increasing in
the utilisation and it has provided the desired results in various fields. The enterprises like the
Google as well as Facebook in United States and the organisation Baidu situated in China has
been using the technology for various purposes. The Google Company utilises the deep
learning methods for executing the efficient image recognition as well as the speech
recognition. In the sector of the image recognition, the deep learning has introduced various
improvements with significant accuracy and it has also observed to have progressed
significantly in variety of the tasks.

4
IMAGE RECOGNITION USING DEEP LEARNING
Discussion
Image classification using deep learning
In the recent years, because of the significantly extensive growth of the digital growth
of the digital content, the automatic classification of the image has grown to be the most
crucial challenge in the indexing of the visual information as well as the retrieval systems
(LeCun, Bengio and Hinton 2015). The computer vision is the interdisciplinary as well as the
subfield of the artificial intelligence that mainly aims in providing the capability of the human
to the computers for comprehending the information from any image. Several efforts has
been made with the help of research for overcoming the issues in the image classification but
the methods mainly considers the low level characteristics of the image primitives
(Goodfellow, Bengio and Courville 2016). With the significant focusing on the low level
image characteristics would not assist with the processing of images. In the recent times, the
image classification has been considered as the major problem within the computer vision. In
the situation of the image understanding, as well as the classification has been done as
significantly easy task, but in the aspect of the computers, it has been considered as a
challenging task (Schmidhuber 2015). Commonly, each image mainly comprises the set of
pixels and then each of the pixel could be represented with distinct values. Moreover, for
storing any image, the computer is required to possess significantly more data. For the
classification of the images, it is required to perform the significantly higher number of
calculations. For efficiently executing the calculations, it is required to have the systems with
significantly higher configurations as well as computing power. In the real time for taking the
decisions based on the input is not possible due to the fact that it requires significantly more
time for executing the computations for providing the results (Deng and Yu 2014).
In the earlier researches, the extraction of features from the Hyper Spectral Images
with the utilisation of the concept of the Convolutional Neural Network of deep learning and

5
IMAGE RECOGNITION USING DEEP LEARNING
it mainly utilises the various layer of pooling in the CNN for the extraction of feature from
HIS that are significantly useful for the perfect classification of the images as well as the
target detection (Liu et al. 2015). It has been observed that it efficiently addresses the
common issues among the features of the HIS. In perspective of the engineering, it seeks in
automating the tasks that could be done by the human visual systems. It has been concerned
with automatic image extraction, the analysis as well as the understanding of the useful
information with the images. In the last decade, various approaches for the image
classification has been described and then compared with various approaches (Dong et al.
2014). But commonly, the image classification denotes to the task of extraction of
information from any image with the labelling of pixels of image to the various classes. In
some of the researches, it has been denoted that the image classification could be executed
with the help of two methods, namely the Supervised classification as well as the
unsupervised classification. The unsupervised classification has been mainly done for the
Unsupervised learning algorithm in the underwater fish recognition framework for the
classification of the images (Sun et al. 2014).
The technique of the unsupervised learning algorithm mainly utilises the pixels of any
image that are clustered into the groups deprived of any intervention of any analyst (Chollet
2017). The grounding on clustered pixels, the information could be retrieved from any image.
In the real world situations, the main availability of the labelled data is significantly less and
therefore the unsupervised classification is mainly done in majority of the situations. The
researchers also discussed the technique of the supervised classification that helps with the
analysis as well as the training of the classifier on labelled image and then extracting the
features from the images. With the proper utilisation of the learned information of training,
newly provided image could be effectively classified on the basis of the features that are
observed in images. In the present times, the algorithms are mainly offering the desired

6
IMAGE RECOGNITION USING DEEP LEARNING
results in areas such as the computer vision (Zhang et al. 2016). Convolutional Neural
Network, that is the machine learning algorithm has been presently utilised for executing the
image classification. In some of the researches, the authors utilises the algorithm of deep
learning for the classification of the quality of the wood board with the effective utilisation of
the texture information from any wood image (Nielsen 2015). It has been inspired with the
biological processes as well as the neurons that are connected in the animal visual cortex.
The convolution layer mainly includes the core building block and it comprises of the
learnable filters as the parameters. Each of the filter has been considered as spatially small
but it extends across depth of input volume (Litjens et al. 2017). The 2-dimensional activation
map has been produced with performing the dot product among the entries as well as the
input of each filter. As result, this network efficiently learns the filter that performs the
activation when it is able to detect any particular characteristic in some of the spatial position
within the input. The pooling layer has been used for executing the sampling of the images
deprived of losing any significant information from images. The max pooling mainly utilises
the maximum value from cluster of the neurons at the prior layer. The completely connected
layers helps in connecting each neuron in any layer to each neuron in any other layer
(Bengio, Goodfellow and Courville 2017). The CNNs mainly utilises the little pre-processing
when matched to the conventional algorithms of classification that utilises the filters that has
been hand engineered. This particular independence of the human intervention in the learning
filters is significantly good advantage of the CNN. The CNN could be considered as the
supervised approach of deep learning that needs significantly large labelled data for executing
the efficient training on network. After training, this model would efficiently learn the main
weights as well as the accuracy of classifier that has been improved (Chetlur et al. 2014). For
example, the self-driving car of Google could be considered as the main project of deep
learning from the Google Company where the image data has been provided as the input

7
IMAGE RECOGNITION USING DEEP LEARNING
from real world situations and then the decisions are made on the basis of the information that
is gained from these images (Long et al. 2015).
Deep learning for the medical image processing
The healthcare sector has been completely dissimilar from other industry. It is on high
priority field and then people consider the significantly high extent of services and care
irrespective of the cost. It has not been able to accomplish the social prospect although it
consumes the significant percentage of the budget (Gupta et al. 2015). Commonly, the
explanations of the medical data is presently been done by any medical expert. In the medical
sector, the accurate diagnosis of any disease mainly relies on the image acquisition as well as
the image interpretations. The devices of the image acquisition has presently improved
significantly and presently the radiological image are available with the significant higher
resolution. It has been presently considered that the medical sector in beginning to gain the
benefits for the automated image interpretation (Gal and Ghahramani 2016). One of most
popular application of machine learning is the computer vision, although the conventional
algorithms of the machine learning that are used for the image interpretations depends
significantly on the expert crafter characteristics. Because of the extensive dissimilarity from
the data of the various patients, the conventional techniques of learning are not dependable.
The machine learning has significantly developed over the last few years with the capability
of shifting through the sophisticated as well as the big data (Papernot et al. 2016).
The technology of deep learning implemented in the medical imaging might become
most disrupting technology that has been introduced in the sector of the radiology since
advent of the digital imaging (Lv et al. 2014). Majority of the researchers considers that the
applications that are based on the deep learning technology would be efficient than the
humans and not solely majority of diagnosis would be executed by the intelligent machines,
but it would also help in the predicting of disease, prescribe the appropriate medicine and also

8
IMAGE RECOGNITION USING DEEP LEARNING
guide in the effective treatment. It has been considered that the sector of Ophthalmology
would be the most primary sector of healthcare that would be significantly revolutionised and
then the cancer diagnosis, pathology would receive the most developments with the
implementation of the deep learning imaging in the healthcare field (Qi et al. 2017). The
vendors and the researchers within the medical sector are progressing in this particular field
with considering the bolder recommendations where the IBM Watson presently boosted the
program with the introduction of the imaging arena by acquisition of the Merge as well as the
Google DeepMind Health. Although several companies are moving towards the introduction
of the deep learning in the various sectors, the future of the deep learning in the medical
imaging has been considered not as effective to any other applications of imaging because of
the complexities included in the field (Chen et al. 2014). This particular notion of the
application of the algorithms based on deep learning for data of medical imaging has been
considered as the fascinating and growing research sector, and there are some sectors that are
helping in bringing down the barriers. These particular challenges has been categorised as the
unavailability of the dataset, the legal and privacy issues, any dedicated medical experts, the
non-standard algorithms of machine learning (Wang, Wang and Yeung 2015).
Several tasks of image diagnosis needs the initial search for identifying the
abnormalities, the quality measurement as well as the changes over significant time. The
automated tool of image analysis based on algorithms of the machine learning are considered
as the key enablers of improving the quality of the image diagnosis as well as the
interpretation by providing the efficient identification of the finding (Ahmed, Jones and
Marks 2015). The deep learning is among the most common methods that is extensively
applied as it provides the required efficiency in the image recognition. It provides significant
methods of analysis of the medical images. The application in the deep learning within the
healthcare covers the broad range of the problems that ranges from disease monitoring to the

9
IMAGE RECOGNITION USING DEEP LEARNING
cancer screening to the personalised suggestions of treatment. Several sources of data that are
available today have introduced immense amount of data at the disposal of the physicians
(Lenz, Lee and Saxena 2015). In the sector of the healthcare, the manual detection process of
Diabetic Retinopathy has been considered as the significant time consuming procedure and
difficult at the existence because of the inaccessibility of the tools as well as expertise.
Application of image analytics for the businesses
The development of the functional as well as accurate technology of image
recognition was considered as the troublesome task, but with introduction of the deep
learning algorithms for the image recognition in the present times, the image recognition has
become significantly simplified. The image recognition application developed with the deep
learning algorithm understands the various patterns that includes the objects in any image,
then recognise the object that has been detected and then analyse the structure of the image.
The image recognition has significant application that creates the value from the perspective
of the business. The functions like the facial recognition, the surveillance as well as security,
the visual geolocation, the object recognition, the gesture recognition, industrial automation,
the code recognition, as well as the medical image analysis are considered as some of the
applications of this technology. With the introduction of this technology in the companies,
the image recognition has started the creation of valuable opportunities of growth in several
fields like the e-commerce, the gaming as well as the automotive industries. Moreover, the
industry where the application of the image recognition is most suitable is the sector of the
social media. Mainly, it is perceived that there are two major methods by which the
companies such as Twitter and Facebook have gained the benefits from the software of image
recognition. The initial benefit is the audience agreement due to the fact that it allows the
audience engagement with the encouraging the users in sharing the images and then tag their
friends, and are able to develop the increasingly connected experience along with driving the

10
IMAGE RECOGNITION USING DEEP LEARNING
social sharing as well as the overall usage of the platforms. The other benefit that could be
gained with the introduction of the image recognition software is the optimisation of
component of mobile advertising of the business.
Conclusion
Therefore, it could be concluded from above discussion that the introduction of the
deep learning algorithm in the image recognition has helped the modern business and
common people with the simplification of image processing. The computer vision is the
interdisciplinary as well as the subfield of the artificial intelligence that mainly aims in
providing the capability of the human to the computers for comprehending the information
from any image. Several efforts has been made with the help of research for overcoming the
issues in the image classification but the methods mainly considers the low level
characteristics of the image primitives. The technique of the unsupervised learning algorithm
mainly utilises the pixels of any image that are clustered into the groups deprived of any
intervention of any analyst. The grounding on clustered pixels, the information could be
retrieved from any image. In real world situations, the main availability of the labelled data is
significantly less and therefore the unsupervised classification is mainly done in majority of
the situations. The researchers also discussed the technique of the supervised classification
that helps with the analysis as well as the training of the classifier on labelled image and then
extracting the features from the images. With the proper utilisation of the learned information
of training, newly provided image could be effectively classified on the basis of the features
that are observed in images. The healthcare sector has been completely dissimilar from other
industry. It is on high priority field and then people consider the significantly high extent of
services and care irrespective of the cost. It has not been able to accomplish the social
prospect although it consumes the significant percentage of the budget. Commonly, the
explanations of the medical data is presently been done by any medical expert.

11
IMAGE RECOGNITION USING DEEP LEARNING
References
Ahmed, E., Jones, M. and Marks, T.K., 2015. An improved deep learning architecture for
person re-identification. In Proceedings of the IEEE conference on computer vision and
pattern recognition (pp. 3908-3916).
Bengio, Y., Goodfellow, I. and Courville, A., 2017. Deep learning (Vol. 1). MIT press.
Chen, Y., Lin, Z., Zhao, X., Wang, G. and Gu, Y., 2014. Deep learning-based classification
of hyperspectral data. IEEE Journal of Selected topics in applied earth observations and
remote sensing, 7(6), pp.2094-2107.
Chetlur, S., Woolley, C., Vandermersch, P., Cohen, J., Tran, J., Catanzaro, B. and Shelhamer,
E., 2014. cudnn: Efficient primitives for deep learning. arXiv preprint arXiv:1410.0759.
Chollet, F., 2017. Xception: Deep learning with depthwise separable convolutions.
In Proceedings of the IEEE conference on computer vision and pattern recognition (pp.
1251-1258).
Deng, L. and Yu, D., 2014. Deep learning: methods and applications. Foundations and
Trends® in Signal Processing, 7(3–4), pp.197-387.
Dong, C., Loy, C.C., He, K. and Tang, X., 2014, September. Learning a deep convolutional
network for image super-resolution. In European conference on computer vision (pp. 184-
199). Springer, Cham.
Gal, Y. and Ghahramani, Z., 2016, June. Dropout as a bayesian approximation: Representing
model uncertainty in deep learning. In international conference on machine learning(pp.
1050-1059).
Goodfellow, I., Bengio, Y. and Courville, A., 2016. Deep learning. MIT press.