International Journal of Scientific Research in Computer Science, Engineering and Information Technology

ISSN : 2456-3307 (www.ijsrcseit.com)

doi : https://doi.org/10.32628/CSEIT206542
Image Processing in Artificial Intelligence
Shahzeb Hussain*1, Prayas Dixit2, Md. Shaayan Hussain3
*1Infosys Limited, Pune, Maharashtra, India
2Infosys Limited, Pune, Maharashtra, India

3Tata Consultancy Services, Chennai, Tamilnadu, India

ABSTRACT

Article Info Machines can learn to elucidate images the same way our brains do and analyse
Volume 6, Issue 5 those images much more thoroughly than we can. When applied to Image
Page Number: 244-249 Processing, Artificial Intelligence (AI) can propel face recognition and security
Publication Issue : functionality in public places, detecting and recognizing intruders, objects, and
September-October-2020 patterns in live images and videos, etc. Image processing technology focuses on
the development of data extraction methods applied to the statistical
classification of visual imagery. In classical image processing systems, an image
is pre-processed to remove noise (denoising), segmented to produce close object
boundaries, analysed to extract a representative feature, and compared to the
ideal object feature vectors by a classifier to decide the nearest object
Article History classification and its associated level. In this paper, we discuss about digital
Accepted : 10 Sep 2020 image processing and the role of AI in it.
Published : 20 Oct 2020 Keywords : Image Processing, Artificial Intelligence, Digital Image, Data
Extraction

I. INTRODUCTION image form. For digital image processing, however,

the output may be an image or some data associated
Basically, image processing is the manipulation of an with that image, such as features, bounding boxes,
image in order to enhance it or extract information characteristics, or masks. In today’s world, image
from it. There are two methods of image processing: processing is widely used in biometrics, gaming,
medical visualization, surveillance, self-driving
• Analog image processing: It is used for processing vehicles, law enforcement, and other fields [1].
physical photographs, printouts, and other hard
copies of images. Here are some of the main objectives of image
• Digital image processing: It is used for processing:
manipulating digital images with the help of
computer algorithms. • Visualization — Represents processed data in an
understandable and more meaningful way, giving
In both cases, we take the input as an image. For visual form to objects that aren’t visible, for
analog image processing, the output is always in an instance.

Copyright: © the author(s), publisher and licensee Technoscience Academy. This is an open-access article distributed
under the terms of the Creative Commons Attribution Non-Commercial License, which permits unrestricted non-
commercial use, distribution, and reproduction in any medium, provided the original work is properly cited
244
 Shahzeb Hussain et al Int J Sci Res CSE & IT, September-October-2020; 6 (5) : 244-249

• Image sharpening and restoration — Improves 2) Image enhancement – This phase enhances the
and enhances the quality of processed images. quality of the captured image to extract hidden
• Image retrieval — Helps in the searching of information from it for further processing.
images.
• Object measurement — Measures objects in an 3) Image restoration – This also improves the quality
image. of an image by removing possible corruptions in
• Pattern recognition — Distinguishes and classifies order to get a clearer version. This process is based
objects in an image, identifies their positions, and mostly on probability and mathematical models and
understands the scene. can be used to get rid of noise, blur, missing pixels,
watermarks, camera misfocus, and other corruptions
that may affect the training of a neural network.

Figure 1: Examples of pattern and object recognition

operations

There are eight key phases included in Image

Processing:

Figure 3: An example of an image with a watermark

4) Color image processing – It includes the processing

of colored images and different color spaces.
Depending on the image type, we can talk about
pseudocolor processing or RGB processing.

5) Image compression and decompression – This

method allows changing of the size and resolution of
an image. Compression is used for reducing the
Figure 2: Key Phases of Digital Image Processing
resolution and size, while decompression is for
restoring an image to its original size and resolution.
Now we will look closer at each of these phases.

These techniques often come into practice during the

1) Image acquisition – It is the process of capturing
image augmentation process. When we have less data,
an image with a sensor (e.g. a camera) and converting
we can extend our dataset with slightly augmented
it into a manageable entity (e.g. a digital image file).
images. In this way, we can improve the way our
One popular image acquisition method is scraping [2].
neural network model generalizes data and make sure
it provides high-quality results [3].

Volume 6, Issue 5, September-October-2020 | http://ijsrcseit.com

245
 Shahzeb Hussain et al Int J Sci Res CSE & IT, September-October-2020; 6 (5) : 244-249

Figure 5: An example of the annotation process of

morphological analysis

Figure 4a: Initial image 7) Image recognition – This is the process of

recognising specific features of particular objects in
an image. Image recognition with AI often utilises
such techniques as object detection, object
recognition, and segmentation. This is where AI
solutions truly flourish. Once we complete these
entire image processing phases, we are ready to build,
train, and test an actual AI solution. The process of
deep learning development includes a full cycle of
operations from data acquisition to incorporating the
developed AI model into the end system [4].

Figure 4b: Augmented images

6) Morphological processing – This method describes Figure 6: Image recognition with a CNN
the shape and structure of the objects in an image. It 8) Representation and description – It is the process
can be used when creating datasets for training AI of visualizing and describing processed images. The
models. In particular, morphological analysis and raw output of an AI system looks like an array of
processing can be applied at the stage where we want numbers and values that represent the data the AI
to describe what we need our AI model to detect or model was trained to produce. A deep neural
recognize. network usually doesn’t produce any output data
representations. With the help of special visualization
tools, these arrays of numbers can be turned into
readable images suitable for further analysis.

Volume 6, Issue 5, September-October-2020 | http://ijsrcseit.com

246
 Shahzeb Hussain et al Int J Sci Res CSE & IT, September-October-2020; 6 (5) : 244-249

The use of Artificial Intelligence and Machine To make it easier to use these methodologies as well
Learning amplifies both the speed of data processing as to implement AI-based image processing
and the quality of the final result. With the aid of AI functionalities in our product, we can use specific
platforms, we can successfully achieve such complex libraries and frameworks. In the next section, we take
tasks as face recognition, object detection, and text a closer look at some of the most popular open-source
recognition. But, in order to get high-quality results, libraries for accomplishing different image processing
we need to pick the right tools and methods [5]. tasks with the help of AI algorithms.

A. OpenCV
II. IMAGE PROCESSING METHODS,
TECHNIQUES, AND TOOLS The Open Source Computer Vision Library (OpenCV)
is a popular library that provides a number of
Most images taken with regular sensors (e.g. camera)
computer and machine learning algorithms and
require pre-processing, as they can contain too much
functions for composing and supporting those
noise. Filtering and edge detection techniques are
algorithms. The library comes with Python, C++, and
two of the most common methods for processing
Java interfaces and supports all popular desktop and
digital images [6].
mobile operating systems. It includes various modules,
• Filtering – This method is used for improving and such as a machine learning module, an image
modifying the input image. With the help of processing module, and an object detection module.
different filters, we can emphasize or remove Using this library, we can acquire, compress, enhance,
certain features in an image, reduce image noise, restore, and extract data from images.
and so on. Popular filtering techniques include
linear filtering, Wiener filtering, and median
B. Visualisation Library
filtering.
• Edge detection – It uses filters for data extraction Visualization Library is C++ middleware for 2D and
and image segmentation. This method helps to 3D applications established on the Open Graphics
find meaningful edges of objects in processed Library. This toolkit allows building portable and
images by detecting discontinuities in brightness. high-performance applications for Windows, Linux,
Roberts edge detection, Sobel edge detection, and and Mac OS systems. It is easy and comfortable to
Canny edge detection are among the most popular work with.
edge detection techniques.
C. VGG Image Annotator
VGG Image Annotator (VIA) is a web-based
application for object annotation. It can be installed
in a web browser and can be used for annotating
detected objects in images, and video records. VIA
doesn’t require additional setup or installation and
can be used with any modern browser.

III.MACHINE LEARNING FRAMEWORKS AND

IMAGE PROCESSING PLATFORMS

If we want to move beyond using simple AI

Figure 7: Examples of edge detection algorithms, we can build custom deep learning

Volume 6, Issue 5, September-October-2020 | http://ijsrcseit.com

247
 Shahzeb Hussain et al Int J Sci Res CSE & IT, September-October-2020; 6 (5) : 244-249

models for image processing. This will make the E. Google Cloud Vision
development a bit faster and easier. We can also use Cloud Vision is a part of the Google Cloud platform
special platforms and frameworks. Now we will take and provides a set of image processing features. It
a look at some of the most popular ones: offers an API for integrating such features as object
A. TensorFlow localization, image labelling and classification, and
object recognition. Cloud Vision uses pre-trained
Google’s TensorFlow is a popular open-source
machine learning models and create and train custom
framework with support of machine learning and
machine learning models for solving different image
deep learning. Using TensorFlow, we can create and
processing tasks [8].
train custom deep learning models. This framework
also provides a set of libraries that can be used in
image processing and computer vision projects. F. Google Colaboratory (Colab)

B. PyTorch Google Colaboratory, known as Colab, is a free cloud

PyTorch is an open-source deep learning framework service that can be used for developing deep learning

created by the Facebook AI Research lab (FAIR). This applications from scratch. Colab makes it easier to use
framework has Python, C++, and Java interfaces. It is popular libraries such as Keras, OpenCV, and

used for building computer vision and natural TensorFlow. Colab provides free GPU resources as

language processing applications. well. In addition to different libraries, frameworks,

and platforms, a large database of images is also
C. MATLAB Image Processing Toolbox
required to train and test our model.
MATLAB is an abbreviation for matrix laboratory.
It’s a popular platform for solving scientific and There are several open databases that contain
mathematical problems. It has an Image Processing millions of tagged images that can be utilised for
Toolbox (IPT) that provides multiple algorithms and training our custom machine learning applications
workflow applications for processing, visualizing, and and algorithms. ImageNet and Pascal VOC are the
analyzing images. MATLAB IPT allows automating most popular.
common image processing workflows. This toolbox
can be used for noise reduction, image segmentation, IV.CONCLUSION
image enhancement, 3D image processing, and other
tasks. A lot of IPT functions support C/C++ code With the help of deep learning algorithms and neural
generation, so they can be utilised for deploying networks, we can teach machines to see and interpret
embedded vision systems and desktop prototyping . [7]
images in the way required for a particular task.
Image processing, itself has brought benefits in
D. Microsoft Computer Vision different areas of technology especially to analyze
images to obtain the necessary information. Progress
Computer Vision is a cloud-based service provided by
in the implementation of AI-based image processing
Microsoft that provides access to advanced algorithms
is impressive and opens a wide range of opportunities
for image processing and data extraction. It allows to:
in fields from medicine and agriculture to retail and
• Moderate image content
law enforcement, geographical remote sensing,
• Extract text from images
robotics, computer and human communication,
• Inspect visual attributes and characteristics of an
healthcare, and satellite communication.
image

Volume 6, Issue 5, September-October-2020 | http://ijsrcseit.com

248
 Shahzeb Hussain et al Int J Sci Res CSE & IT, September-October-2020; 6 (5) : 244-249

V. REFERENCES Cite this article as :

[1] Patel, Krishna Kumar, A. Kar, S. N. Jha, and M. Shahzeb Hussain, Prayas Dixit, Md. Shaayan Hussain,
A. Khan. "Machine vision system: a tool for "Image Processing in Artificial Intelligence",
quality inspection of food and agricultural International Journal of Scientific Research in
products." Journal of food science and technology Computer Science, Engineering and Information
49, no. 2 (2012): 123-141. doi: 10.1007/s13197- Technology (IJSRCSEIT), ISSN : 2456-3307, Volume
011-0321-4 6 Issue 5, pp. 244-249, September-October 2020.
[2] Long, Jonathan, Evan Shelhamer, and Trevor Available at
Darrell. "Fully convolutional networks for doi : https://doi.org/10.32628/CSEIT206542
semantic segmentation." In Proceedings of the Journal URL : http://ijsrcseit.com/CSEIT206542
IEEE Conference on Computer Vision and
Pattern Recognition, pp. 3431-3440. 2015. DOI:
10.1109/CVPR.2015.7298965
[3] K. D. Irwin and G. C. Hilton, "Transition-Edge
Sensors," Cryogenic Particle Detection, ed. C.
Enss, Springer, 2005
[4] R. Deriche, Using Canny's criteria to derive a
"recursively implemented optimal edge detector",
Int. J. Computer Vision, Vol. 1, pp. 167-187, April
1987.
[5] Rautaray, Siddharth S., and Anupam Agrawal.
"Vision-based hand gesture recognition for
human-computer interaction: a survey." Artificial
Intelligence Review 43, no. 1 (2015): 1-54. Doi:
10.1007/s10462-012-9356-9
[6] "NDT Resource Center. " http://www.ndt-ed.org
/index_flash.htm.
[7] "Image processing toolbox user's guide." (2005)
The Math Works, Natick, Massachusetts, USA.
[8] Kehoe, Ben, Sachin Patil, Pieter Abbeel, and Ken
Goldberg. "A survey of research on cloud robotics
and automation." IEEE Transactions on
automation science and engineering 12, no. 2
(2015): 398-409. DOI:
10.1109/TASE.2014.2376492

Volume 6, Issue 5, September-October-2020 | http://ijsrcseit.com

249